TITLE
Galley: Modern Query Optimization for Declarative Sparse Tensor Programming

ABSTRACT
Modern computation has become dominated by the tensor programming abstraction. This framework allows users to write high performance programs for bulk computation via a high-level imperative interface. Recent work has extended this paradigm to sparse tensors (i.e. tensors where most entries are not explicitly represented) with the use of sparse tensor compilers. These systems excel at producing efficient code for computation over sparse tensors, which may be stored in a wide variety of formats. However, they require the user to manually choose the order of operations and the data formats at every step. Unfortunately, these decisions are both highly impactful and complicated, requiring significant effort to manually optimize. In this talk, I'll present Galley, a system for declarative sparse tensor programming. Galley adapts the database community's expertise in cost-based optimization to sparse tensor programs. First, it synthesizes a logical plan which determines high level structure and materialization strategy. Then, it lowers each piece of the logical plan to a physical implementation by determining loop orders, tensor formats, and merge algorithms. This plan is then efficiently executed by a sparse tensor compiler. Lastly, I'll present early experimental results which show how Galley can produce up to 100x speedups over hand-optimized implementations for modestly complex tensor programs.

RELATED WORK:
* Galley: Modern Query Optimization for Sparse Tensor Programs. 2024. https://arxiv.org/pdf/2408.14706

BIO
Kyle Deeds is currently a 5th year PhD student in the database lab at the University of Washington. He is advised by Dan Suciu and Magda Balazinska and interested in the intersection of databases, compilers, and HPC. Broadly, his work aims to make high performance computing accessible to domain experts through declarative programming coupled with automatic program optimization.

TITLE
Towards the Efficiency, Robustness and Trustworthiness of AI Models

ABSTRACT
Deep learning models have demonstrated remarkable success across diverse domains. However, improvements are needed towards the efficiency, robustness and trustworthiness of these models in order to deploy them to real-world applications. In this talk, I will discuss our efforts towards improving these three important dimensions of AI models. First, I will present efficient deep learning models that we developed for resource-constrained scenarios. We demonstrate that these lightweight models require much less computational time and storage without much accuracy drop. Second, I will present the methodologies we developed to systematically create adversarial samples that could skew the model's original objectives for a series of machine learning tasks. These efforts contribute to a better understanding of the deep learning models. The last part of the talk will focus on our recent efforts towards evaluating and improving the trustworthiness and reliability of large language model output.

RELATED WORK:
-  “Data Poisoning Attacks Against Outcome Interpretations of Predictive Models”, KDD 2021. https://dl.acm.org/doi/10.1145/3447548.3467405
-  “LOKI: A Practical Data Poisoning Attack Framework Against Next Item Recommendations”, TKDE 2023. https://ieeexplore.ieee.org/document/9806383
- “LightToken: a Task and Model-agnostic Lightweight Token Embedding Framework for Pre-trained Language Models”, KDD’23. https://dl.acm.org/doi/abs/10.1145/3580305.3599416
- “HadSkip: Homotopic and Adaptive Layer Skipping of Pre-trained Language Models for Efficient Inference”, EMNLP’23 Findings. https://aclanthology.org/2023.findings-emnlp.283/
- “Row and Column-wise Sparse Low-rank Adaptation of Pre-trained Language Model for Knowledge Editing and Fine-tuning”. https://arxiv.org/abs/2406.10777

 The following two papers will be briefly mentioned in the talk:

- “Evaluating the Factuality of Large Language Models using Large-Scale Knowledge Graphs”, https://arxiv.org/abs/2404.00942

- “SaySelf: Teaching LLMs to Express Confidence with Self-Reflective Rationales”, https://arxiv.org/abs/2405.20974

BIO

Jing Gao is an Associate Professor in the Elmore Family School of Electrical and Computer Engineering, Purdue University. Before joining Purdue in January 2021, she was an Associate Professor in the Department of Computer Science and Engineering at the University at Buffalo (UB), State University of New York. She received her PhD from Computer Science Department, University of Illinois at Urbana Champaign in 2011, and subsequently joined UB in 2012. She is broadly interested in data and information analysis with a focus on data mining. In particular, she is interested in information veracity analysis, multi-source data analysis, knowledge graphs, text mining, data and model efficiency, fairness and interpretation, transfer learning, federated learning, crowdsourcing, data stream mining, and anomaly detection. She has published over 200 papers in referred journals and conferences. Her publications have received over 19,000 citations and her H-index is 68. She is an editor of ACM Transactions on Intelligence Systems and Technology (TIST) and IEEE Transactions on Knowledge and Data Engineering (TKDE). She serves as the Program Co-Chair of the 2024 SIAM Conference on Data Mining, Web Mining Track Co-Chair of the 2024 Web Conference, and a senior program committee member of many leading conferences in data science. She is a recipient of NSF CAREER award, IBM faculty award, ICDM Tao Li Award, SDM/IBM Early Career Award and UIUC CS Early Career Academic Achievement Alumni Award.

TITLE
Longshot: Indexing Growing Databases using MPC and Differential Privacy
 
ABSTRACT
In this work, we propose Longshot, a novel design for secure outsourced database systems that supports ad-hoc queries through the use of secure multi-party computation and differential privacy. By combining these two techniques, we build and maintain data structures (i.e., synopses, indexes, and stores) that improve query execution efficiency while maintaining strong privacy and security guarantees. As new data records are uploaded by data owners, these data structures are continually updated by Longshot using novel algorithms that leverage bounded information leakage to minimize the use of expensive cryptographic protocols. Furthermore, Longshot organizes the data structures as a hierarchical tree based on when the update occurred, allowing for update strategies that provide logarithmic error over time. Through this approach, Longshot introduces a tunable three-way trade-off between privacy, accuracy, and efficiency. Our experimental results confirm that our optimizations are not only asymptotic improvements but also observable in practice. In particular, we see a 5x efficiency improvement to update our data structures even when the number of updates is less than 200. Moreover, the data structures significantly improve query runtimes over time, about ∼10^3x faster compared to the baseline after 20 updates.


RELATED WORK:
- Shrinkwrap: Efficient SQL Query Processing in Differentially Private Data Federations. PVLB 2018.https://dl.acm.org/doi/10.14778/3291264.3291274
- Cryptϵ: Crypto-Assisted Differential Privacy on Untrusted Servers. SIGMOD 2020. https://dl.acm.org/doi/10.1145/3318464.3380596
BIO
Johes Bater is an assistant professor of Computer Science at Tufts University. Before that, he was a postdoctoral researcher in the Database Group at Duke University, and received his Ph.D. in computer science from Northwestern University, and completed his B.S. and M.S. in electrical engineering at Stanford University. His research centers on how to balance privacy, security, and utility to build fast, accurate database systems that support privacy-preserving analytics with provable security guarantees.

TITLE
Circuits for Sum-Product Queries
 
ABSTRACT
In this talk, we ask the following question: given a Boolean Conjunctive Query, what is the smallest circuit that computes the provenance polynomial of the query over a given semiring? We answer this question by giving upper and lower bounds. We show a circuit construction that matches this bound when the semiring is idempotent. The techniques we use combine several central notions in database theory: provenance polynomials, tree decompositions, and disjunctive Datalog programs. We extend our results to lower and upper bounds for formulas and to bounds for non-Boolean CQs.
 
RELATED PAPER
Tight Bounds of Circuits for Sum-Product Queries. Austen Z. Fan, Paraschos Koutris, and Hangdong Zhao. PODS 2024. https://doi.org/10.1145/3651588


BIO
Paris Koutris is an associate professor in Computer Sciences at the University of Wisconsin-Madison. His research lies in the intersection of data management theory and practice, focusing on data processing for massively parallel systems, data markets, efficient data structures, and managing uncertain data. He has won the SIGMOD Jim Gray Dissertation Award for his work on the foundations of parallel data processing and has also won the 2023 PODS Alberto O. Mendelzon Test-of-Time award. 

TITLE
Optimally Rewriting Formulas and Database Queries: A Confluence of Term Rewriting, Structural Decomposition, and Complexity

ABSTRACT
A central computational task in database theory, finite model theory, and computer science at large is the evaluation of a first-order sentence on a finite structure.  In the context of this task, the width of a sentence, defined as the maximum number of free variables over all subformulas, has been established as a crucial measure, where minimizing width of a sentence (while retaining logical equivalence) is considered highly desirable.
 
An undecidability result rules out the possibility of an algorithm that, given a first-order sentence, returns a logically equivalent sentence of minimum width; this result motivates the study of width minimization via syntactic rewriting rules, which is this article's focus.  For a number of common rewriting rules (which are known to preserve logical equivalence), including rules that allow for the movement of quantifiers, we present an algorithm that, given a positive first-order sentence \phi, outputs the minimum-width sentence obtainable from \phi via application of these rules.  We thus obtain a complete algorithmic understanding of width minimization up to the studied rules; this result is the first one---of which we are aware---that establishes this type of understanding in such a general setting.  Our result builds on the theory of term rewriting and establishes an interface among this theory, query evaluation, and structural decomposition theory.

RELATED WORK
* Chen, Hubie, and Stefan Mengel. "Optimally Rewriting Formulas and Database Queries: A Confluence of Term Rewriting, Structural Decomposition, and Complexity." 27th International Conference on Database Theory (ICDT 2024). Schloss Dagstuhl–Leibniz-Zentrum für Informatik, 2024. https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ICDT.2024.16

SHORT BIO
Hubie Chen is an academic in the Department of Informatics at King’s College London.  Prior to working in the UK, he was an Ikerbasque Research Professor in Donostia-San Sebastian, Spain.  He has held invited positions at institutions such as E ́cole Polytechnique, Humboldt-Universit ̈at zu Berlin, and Universit ̈at Wien. He received PhD and BS degrees in computer science and mathematics from Cornell University and Stanford University, respectively.  He is author of the MIT Press textbook "Computability and Complexity", a state-of-the-art introduction to computability theory and complexity theory.
His current research interests include the theoretical foundations of database query evaluation and database query optimization, and he is broadly interested in computational complexity, logic, database theory, and a number of related areas.

TITLE
Mosaic: An Architecture for Scalable & Interoperable Data Views – Dominik’s Version

ABSTRACT
Mosaic is an architecture for greater scalability, extensibility, and interoperability of interactive data views. Mosaic decouples data processing from specification logic: clients publish their data needs as declarative queries that are then managed and automatically optimized by a coordinator that proxies access to a scalable data store. Mosaic generalizes Vega-Lite’s selection abstraction to enable rich integration and linking across visualizations and components such as menus, text search, and tables. In this talk, I will demonstrate Mosaic’s expressiveness, extensibility, and interoperability through examples that compose diverse visualization, interaction, and optimization techniques—many constructed using vgplot, a grammar of interactive graphics in which graphical marks act as Mosaic clients. Benchmarks show order-of-magnitude performance improvements over existing web-based visualization systems—enabling flexible, real-time visual exploration of billion+ record datasets. I’ll conclude by discussing Mosaic’s potential as an open platform that bridges visualization languages, scalable visualization, and interactive data systems more broadly.
 
RELATED WORK
Heer, Jeffrey, and Dominik Moritz. "Mosaic: An architecture for scalable & interoperable data views." IEEE Transactions on Visualization and Computer Graphics (2023). https://doi.org/10.1109/tvcg.2023.3327189

SHORT BIO
Dominik Moritz is on the faculty at Carnegie Mellon University where he co-directs the Data Interaction Group (https://dig.cmu.edu/) at the Human-Computer Interaction Institute. His group’s research develops interactive systems that empower everyone to effectively analyze and communicate data. Dominik also manages the visualization team in Apple’s machine learning organization. His systems (Vega-Lite, Falcon, Draco, Voyager, and others) have won awards at academic venues (e.g. IEEE VIS and CHI), are widely used in industry, and by the Python and JavaScript data science communities. Dominik got his PhD from the Paul G. Allen School at the University of Washington, where he was advised by Jeff Heer and Bill Howe. He recently received the significant new researcher award at IEEE VIS 2024.

TITLE
Mitigating Filter Bubbles Using Diffusion Models

ABSTRACT
While social networks greatly facilitate information dissemination, they are well known to have contributed to the phenomena of filter bubbles and echo chambers. This in turn can lead to societal polarization and erosion of trust in public institutions. Mitigating filter bubbles is an urgent open problem.

Recently, approaches based on the influence maximization paradigm have been explored for mitigating filter bubbles by balancing exposure to opposing viewpoints. However, existing works ignore the inherent competition between the adoption of opposing viewpoints by users.

In this talk, I will describe a realistic model for the filter bubble problem, which unlike previous work, captures the  competition between opposing opinions propagating in a network as well as the complementary nature of the reward for exposing users to both those opinions. We formulate an optimization problem for mitigating filter bubbles under this model. We establish several evidences of the intrinsic difficulty in developing constant approximation to the problem and develop a heuristic and two instance-dependent approximation algorithms. Our experiments over 4 real datasets show that our heuristic far outperforms two state-of-the-art baselines as well as other algorithms in both efficiency and mitigating filter bubbles. We also empirically demonstrate that our best heuristic performs close to the optimal objective, which is verified by utilizing the theoretical bounds of our approximation algorithms.
 
RELATED WORK

- Banerjee, Chen, and L. Mitigating Filter Bubbles Under a Competitive Diffusion Model. SIGMOD 2023 https://dl.acm.org/doi/10.1145/3589320
- Banerjee, Chen, and L. Maximizing social welfare in a competitive diffusion model. VLDB 2020 https://dl.acm.org/doi/abs/10.14778/3436905.3436920
- Lu, Chen, and L. From Competition to Complementarity: Comparative Influence Diffusion and Maximization. VLDB 2016 https://dl.acm.org/doi/10.14778/2850578.2850581

SHORT BIO 
Laks V.S. Lakshmanan is a professor of Computer science at UBC, Vancouver, Canada. His research interests span a wide spectrum of topics in data management, integration, cleaning, and warehousing; data mining; semi-structured and unstructured data; big graphs, social networks and social media; ML, NLP; and efficient deep learning. He is an ACM Distinguished Scientist and has won several awards including best paper awards and distinguished reviewer awards. He has served on most top conferences and journals in his areas of research, on program committees, as senior PC member, meta-reviewer, general chair, and as associate editor.

TITLE
The Case for Machine Learning over Relational Data

ABSTRACT
As we witness the data science revolution, each research community legitimately reflects on its relevance and place in this new landscape. The database research community has at least three reasons to feel empowered by this revolution. This has to do with the pervasiveness of relational data in data science, the widespread need for efficient data processing, and the new processing challenges posed by data science workloads beyond the classical database workloads. The first two aforementioned reasons are widely acknowledged as core to the community’s raison d’être. The third reason explains the longevity of relational database management systems success: Whenever a new promising data-centric technology surfaces, research is under way to show that it can be captured naturally by variations or extensions of the existing relational techniques.

In this talk, I will make the case for a first-principles approach to machine learning over relational databases that guided recent development in database systems and theory. This includes theoretical development on the algebraic and combinatorial structure of relational data processing. It also includes systems development on compilation for hybrid database and learning workloads and on computation sharing across aggregates in learning-specific batches. Such development can dramatically boost the performance of machine learning.

This work is the outcome of extensive collaboration of the author with colleagues from RelationalAI (https://www.relational.ai) and the FDB research project (https://fdbresearch.github.io), in particular: Mahmoud Abo Khamis, Molham Aref, Ahmet Kara, Hung Ngo, XuanLong Nguyen, Milos Nikolic, Maximilian Schleich, Amir Shaikhha, and Haozhe Zhang.

RELATED WORK
1. Dan Olteanu: The Relational Data Borg is Learning. PVLDB 2020. http://www.vldb.org/pvldb/vol13/p3502-olteanu.pdf

2. Mahmoud Abo Khamis, Hung Q. Ngo, XuanLong Nguyen, Dan Olteanu, Maximilian Schleich:
Learning Models over Relational Data Using Sparse Tensors and Functional Dependencies. TODS 2020. https://dl.acm.org/doi/pdf/10.1145/3375661

3. Mahmoud Abo Khamis, Ryan R. Curtin, Benjamin Moseley, Hung Q. Ngo, XuanLong Nguyen, Dan Olteanu, Maximilian Schleich:
Functional Aggregate Queries with Additive Inequalities. TODS 2020. https://dl.acm.org/doi/pdf/10.1145/3426865

BIO
Dan Olteanu is a professor at the University of Zurich, where he leads the Data Systems and Theory group (https://www.ifi.uzh.ch/en/dast.html), and a computer scientist at RelationalAI (https://relational.ai). He currently works on incremental view maintenance, cardinality estimation, in-database machine learning and linear algebra, adaptive query processing, and fact attribution in query answering.

TITLE
Returning paths in Graph Database Queries

ABSTRACT
In this talk we describe a theoretical framework for returning paths in graph database queries. The work subsumes the path semantics supported in the latest Graph Query Language (GQL) ISO standard and provides a platform for extending the approach for more expressive queries. We also discuss what challenges this approach has in terms of implementation and why it scales remarkably well over real world knowledge graphs, despite the theory telling us that this should not be the case. Overall, the talk should give you a good understanding of why path queries work in practice and why converting the theory into working implementations is a very useful exercise.

RELATED WORK
1. Benjamín Farías, Wim Martens, Carlos Rojas, Domagoj Vrgoč. PathFinder: A unified approach for handling paths in graph query languages. Arxiv 2023. https://arxiv.org/abs/2306.02194
2. Pathfinder live Demo: https://mdb.imfd.cl/path_finder/

BIO

Domagoj Vrgoč is an Associate Professor at  Pontificia Universidad Católica de Chile, and an Associate Researcher at the Millennium Institute for Foundational Research on Data (IMFD). He  has over a decade of research experience in the area of graph query languages and algorithms. He was invited to deliver several tutorial presentations, including SIGMOD'24, ISWC'24, AMW'23, SPIRE'22, and he received several awards for his work, including ICDT Test-of-time award in 2023, and SIGMOD best industry paper award in 2023. He is the project lead of the open source graph engine MillenniumDB.

TITLE
Einsummable: Towards Fully Automatic Programming Systems for ML

ABSTRACT
Training and inference for very large ML models on large compute clusters is difficult. It’s even harder to get a very large model to work on a small compute cluster with limited memory. A programmer must understand how to break up a model so that it can be distributed to many machines or paged in and out in the case of limited memory. This is difficult or even impossible for most mere mortals. I’ll talk about how adapting ideas from relational database system design can result in ML systems that automatically enable training and inference for almost any model on almost any hardware, with little expertise on the part of the user.

RELATED PAPERS
* EinDecomp: Decomposition of Declaratively-Specified Machine Learning and Numerical Computations for Parallel Execution
https://arxiv.org/abs/2410.02682
* TURNIP: A "Nondeterministic" GPU Runtime with CPU RAM Offload
https://arxiv.org/abs/2405.16283

BIO
Christopher Jermaine is the chair of the Rice University Department of Computer Science in the George R. Brown School of Engineering. He is also the program director of the Master of Data Science program, professor of Computer Science and J.S. Abercrombie Professor of Engineering.

TITLE
Graph Pattern Matching in GQL and SQL/PGQ

ABSTRACT
GQL (Graph Query Language) is being developed as a new ISO standard for graph query languages to play the same role for graph databases as SQL plays for relational databases. In parallel, an extension of SQL for querying property graphs, SQL/PGQ, has recently been added to the SQL standard; it shares the graph pattern matching functionality with GQL. Both standards are hard-to-understand specifications of hundreds of pages. The goal is to present a digest of the language that is easy for the research community to understand, and thus to initiate research on these future standards for querying graphs. The focus will be on pattern matching features shared by GQL and SQL/PGQ, as well as querying facilities of GQL.

RELATED PAPERS
* Nadime Francis, Amélie Gheerbrant, Paolo Guagliardo, Leonid Libkin, Victor Marsault, Wim Martens, Filip Murlak, Liat Peterfreund, Alexandra Rogova, and Domagoj Vrgoč. A Researcher’s Digest of GQL (Invited Talk). In 26th International Conference on Database Theory (ICDT 2023). https://doi.org/10.4230/LIPIcs.ICDT.2023.1 
*  Alin Deutsch, Nadime Francis, Alastair Green, Keith Hare, Bei Li, Leonid Libkin, Tobias Lindaaker, Victor Marsault, Wim Martens, Jan Michels, Filip Murlak, Stefan Plantikow, Petra Selmer, Oskar van Rest, Hannes Voigt, Domagoj Vrgoč, Mingxi Wu, and Fred Zemke. 2022. Graph Pattern Matching in GQL and SQL/PGQ. In Proceedings of the 2022 International Conference on Management of Data (SIGMOD '22). https://doi.org/10.1145/3514221.3526057

BIO

Wim Martens (University of Bayreuth, Germany) is interested in foundational aspects of data management, with a current focus on query language design, graph databases, logic, complexity, and formal language theory. One of his goals is to find a good balance for making programming with data easy, intuitive, precise, and efficient. He was an invited speaker at PODS 2022 and STOC 2017 and his research received several awards at leading conferences such as SIGMOD, ICDT, and ISWC. Since 2022, he has been working with RelationalAI on foundations of relational programming languages.

TITLE
What Teaching Databases Taught me about Researching Databases

ABSTRACT
Declarative querying and automatic optimization are the cornerstones of the success and longevity of database systems, yet these concepts often pose challenges for novice learners accustomed to different coding paradigms. The transition is further hampered by the lack of query debugging tools (be it for correctness or performance) compared to the plethora available for programming languages. The talk samples several systems that we build at Duke University to help students learn and debug database queries. These systems have not only helped scale up teaching and improve learning, but also inspired research on interesting and fundamental questions concerning databases. Furthermore, with the rise of generative AI, we argue that there is a heightened need for skills in scrutinizing and debugging AI-generated queries, and we outline several ongoing and future work directions aimed at addressing this emerging challenge.

RELATED PAPERS
* Hu, Gilad, Stephens-Martinez, Roy, Yang. Qr-Hint: Actionable Hints Towards Correcting Wrong SQL Queries. SIGMOD 2024 https://arxiv.org/abs/2404.04352
* Miao, Roy, Yang. Explaining Wrong Queries Using Small Examples. SIGMOD 2019 https://dl.acm.org/doi/10.1145/3299869.3319866
* Gilad, Miao, Roy, Yang. Understanding Queries by Conditional Instances. SIGMOD 2022 https://dl.acm.org/doi/10.1145/3514221.3517898

BIO
Jun Yang is currently the Bishop-MacDermott Family Professor of Computer Science at Duke University. He joined Duke after receiving his Ph.D. from Stanford in 2001 and chaired the Department of Computer Science at Duke during 2020-2023. He has broad research interests in databases and data-intensive systems. He is a Trustee of the VLDB Endowment and served as the general co-chair of SIGMOD 2017 and the co-Editor-in-Chief of PVLDB during 2022-2023. He is a recipient of the CAREER Award, IBM Faculty Award, HP Labs Innovation Research Award, and Google Faculty Research Award. He has striven to connect research to his other passions, such as journalism, where he has worked on computational fact-checking since its nascent days, and education, where he has built a number of software tools for learning databases. He received the David and Janet Vaughan Brooks Teaching Award at Duke.
https://users.cs.duke.edu/~junyang/

TITLE
Two-Commodity Flow is Equivalent to Linear Programming

ABSTRACT
Maximum flow is a very well-studied problem, which can be formulated as a linear program. While an almost-linear time algorithm for single-commodity flow is developed, progress for multi-commodity flow is only made by progress in general linear program solvers. It turns out it is not by accident since two-commodity flow problem (2CF) is equivalent to general linear programs (LP). More precisely, we show that any LP (under milder assumptions) can be encoded as a 2CF problem in nearly linear time, with only a polylogarithmic blow-up in problem size and polynomial blow-up in error. Our reduction applies to high-accuracy approximation algorithms and exact algorithms.
Our proof builds upon Itai's polynomial-time reduction from LP to 2CF [Itai JACM'78], improving its reduction time, problem size, and establishing an error bound. In this talk, I will go through the key steps of the reduction algorithm from LP to 2CF. Additionally, I will provide an intuitive error analysis of the solution mapping from a 2CF to LP.
Joint work with Rasmus Kyng and Peng Zhang.

RELATED WORK
* Ming, Kyng, Zhang. Two-commodity flow is equivalent to linear programming under nearly-linear time reductions, 49th EATCS International Conference on Automata, Languages, and Programming, https://arxiv.org/abs/2201.11587
* Presentation at Hausdorff Center for Mathematics: Two-Commodity Flow is as Hard as Linear Programming https://www.youtube.com/watch?v=f6SWJMQIJRc
* Alon Itai. Two-commodity flow. JACM 1978. https://dl.acm.org/doi/10.1145/322092.322100

SHORT BIO
I am a fourth-year PhD student at the Department of Computer Science, ETH Zürich. I am fortunate to be supervised by Rasmus Kyng. My research interests lie in algorithms and optimization problems, with a specific focus on fast algorithms and complexity analysis for structured Linear Equations and Linear Programming. Before coming to ETH Zürich, I received my Bachelor’s degree from Shanghai Jiao Tong University. Additionally, I participated in a dual-degree program, obtaining a Diplôme d’Ingénieur from CentraleSupélec in France and a Master’s degree from Shanghai Jiao Tong University in China.
https://2015dingm.github.io/

TITLE
Tractability beyond beta-acyclicity for conjunctive queries with negation and SAT

ABSTRACT
Numerous fundamental database and reasoning problems are known to be NP-hard in general but tractable on instances where the underlying hypergraph structure is β-acyclic. Despite the importance of many of these problems, there has been little success in generalizing these results beyond acyclicity.

In this paper, we take on this challenge and propose nest-set width, a novel generalization of hypergraph β-acyclicity. We demonstrate that nest-set width has desirable properties and algorithmic significance. In particular, evaluation of boolean conjunctive queries with negation (CQ ¬ ) is tractable for classes with bounded nest-set width. Furthermore, propositional satisfiability (SAT) is fixed-parameter tractable when parameterized by nestset width.

RELATED WORK
* Matthias Lanzinger. Tractability beyond β-acyclicity for conjunctive queries with negation and SAT. Theoretical Computer Science, 2023.
https://doi.org/10.1016/j.tcs.2022.12.002

SHORT BIO
Matthias Lanzinger is currently an Assistent Professor (Tenure Track) for Database Theory at the Databases and Artificial Intelligence Group. He obtained his PhD from TU Vienna in 2021 and afterwards worked as a Senior Research Assoicate at the University of Oxford until the end of 2023.
https://dbai.tuwien.ac.at/staff/mlanzing/

TITLE
Parallelizing GPU-based Mini-Batch Graph Neural Network Training

ABSTRACT
Many datasets are best represented as graphs of entities connected by relationships rather than as a single uniform dataset or table. Graph Neural Networks (GNNs) have been used to achieve state-of-the-art performance in tasks such as node classification and link prediction. This talk will discuss recent advances in parallelizing GPU-based mini-batch GNN training.

The focus will be on the widely used mini-batch approach to GNN training, which involves three main steps in each iteration: sampling the k-hop neighbors of the mini-batch, loading the samples onto the GPUs, and performing the training. The first part of the talk will discuss NextDoor, a GPU-based sampling system that showed the importance of GPU-based sampling to speed up end-to-end GNN training. NextDoor introduces a new form of parallelism, called transit parallelism, to maximize GPU resource utilization and make sampling faster. The second part of the talk will introduce split parallelism, a novel approach to parallel GNN training targeting the entire end-to-end training pipeline. It will present GSplit, a split-parallel system that avoids redundant data loads and makes multiple GPUs sample and train cooperatively. Finally, the talk will discuss the results of an experimental comparison between full-graph and mini-batch training systems.

RELATED WORK:
* Accelerating Graph Sampling for Graph Machine Learning using GPUs, EuroSys 2021. https://marcoserafini.github.io/assets/pdf/nextdoor.pdf
* GSplit: Scaling Graph Neural Network Training on Large Graphs via Split-Parallelism, arxiv preprint 2024. https://arxiv.org/pdf/2303.13775
* Graph Neural Network Training Systems: A Performance Comparison of Full-Graph and Mini-Batch, arxiv preprint 2024. https://arxiv.org/pdf/2406.00552

BIO
Marco Serafini is an assistant professor at the Manning College of Information and Computer Sciences at UMass Amherst. He works on systems for graph learning, mining, and data management (e.g., the Arabesque, LiveGraph, NextDoor, and GSplit projects), and cloud data management systems, (e.g., Accordion, E-Store, and Clay). His previous work contributed to the Apache Zookeeper and Storm projects and won a test-of-time award at DSN 2021. He was in the Program Committees of major conferences in systems and database management, including SOSP, OSDI, Eurosys, SIGMOD, ASPLOS, VLDB, and ICDE. He was the Program Chair of the APSys and LADIS workshops and an AE for SIGMOD.

TITLE
A New Generation of Query Processing for Advanced Data Analytics

ABSTRACT
While both the amount of data and our need to analyze them are growing, today’s data systems are limited in the analytical tasks they can handle efficiently. When presented with a complex analytical task, they often become unresponsive because they resort to naive query-processing algorithms. A major culprit is a join operation in the middle of the query-processing pipeline: that join can often produce unnecessarily large intermediate results even though the final output is small. Overcoming this challenge by “pushing” different kinds of operators behind joins is not always simple and often requires advancements in algorithms.

In this talk, I will present my work on improving the algorithmic capabilities of data systems, focused on the case where users have preferences over the query answers and are interested in retrieving only a select few of them. I show that for a large class of these queries, it is possible to avoid materializing intermediate joins, resulting in non-trivial complexity guarantees and efficient implementations that outperform existing systems by orders of magnitude. Besides addressing fundamental questions regarding the limits of query processing, this work opens up unexplored possibilities for system design. The long-term goal is a new generation of scalable systems for data analytics that can efficiently handle queries and data that are considered infeasible today.

BIO
Nikolaos (Nikos) Tziavelis is a PhD candidate at Northeastern University, advised by Mirek Riedewald and Wolfgang Gatterbauer, and before that, he received a Diploma in Electrical and Computer Engineering from the National Technical University of Athens. His research interests lie in the design and implementation of novel practical algorithms for query processing in data systems. His work has been awarded with a Google PhD fellowship, a PODS 2021 “Best of” recognition, and a 2023 PhD Workshop Best Paper Award.
https://ntzia.github.io/

Title: Machine Learning for Databases: Foundations, Paradigms, and Open problems

Abstract: The database community has actively explored how machine learning (ML) can be leveraged to address challenges in database systems (ML4DB) for the past several years. The talk will briefly cover several selected ML4DB tasks, and then focus on our perceptions on foundations and paradigms. First, query plan is used as input in several ML4DB tasks, and most studies usually focus on one task and develop a new design to represent query plans along with a ML4DB framework. This talk will present a different approach to these tasks by abstracting the query plan representation as a common component or foundation for multiple ML4DB tasks. This talk will also present a summary of finding of experimental study on query plan representation. Second, learned indexes are proposed to learn a ML model to replace the classic indexes. This talk will present a different paradigm of using ML for indexes, namely enhancing the classic indexes with ML, rather than replacing them. The talk will end with a discussion on some open problems.

Several related papers:

• Yue Zhao, Gao Cong, Jiachen Shi, Chunyan Miao: QueryFormer: A Tree Transformer Model for Query Plan Representation. Proc. VLDB Endow. 15(8): 1658-1670 (2022)
• Yue Zhao, Zhaodonghui Li, Gao Cong: A Comparative Study and Component Analysis of Query Plan Representation Techniques in ML4DB Studies, VLDB 24
• Jingyi Yang, Gao Cong: PLATON: Top-down R-tree Packing with Learned Partition Policy. SIGMOD24
• Tu Gu, Kaiyu Feng, Gao Cong, Cheng Long, Zheng Wang, Sheng Wang: The RLR-Tree: A Reinforcement Learning Based R-Tree for Spatial Data. SIGMOD 2023
• Jiangneng Li, Zheng Wang, Gao Cong, Cheng Long, Han Mao Kiah, Bin Cui: Towards Designing and Learning Piecewise Space-Filling Curves. Proc. VLDB Endow. 16(9): 2158-2171 (2023)

Bio: Gao Cong is currently a Professor in the School of Computer Science and Engineering at Nanyang Technological University (NTU), and a co-director for Singtel Cognitive and Artificial Intelligence Lab for Enterprises@NTU. He previously worked at Aalborg University, Denmark, Microsoft Research Asia, and the University of Edinburgh. His current research interests include ML4DB, spatial data management, spatial-temporal data mining, and recommendation systems. He received the best paper runner-up awards at the WSDM'20 and WSDM'22 conferences for two of his research papers. His citation in Google Scholar was over 18,000 with H-index 70. He served as a PC co-chair for ICDE’2022, the associate general chair of KDD’21, a PC co-chair for E&A track of VLDB 2014, and a PC vice-Chair for ICDE’18. He is an associate editor for ACM Transactions on Database Systems (TODS) and the vice Chair of ACM KDD Singapore chapter.

TITLE
The Complexity of Resilience Problems via Valued Constraint Satisfaction Problems

ABSTRACT
Valued constraint satisfaction problems (VCSPs) constitute a large class of computational optimisation problems. It was shown recently that, over finite domains, every VCSP is in P or NP-complete, depending on the admitted cost functions. In this article, we study cost functions over countably infinite domains whose automorphisms form an oligomorphic permutation group. Our results include a hardness condition based on a generalisation of pp-constructability as known from classical CSPs and a polynomial-time tractability condition based on the concept of fractional polymorphisms. We then observe that the resilience problem for unions of conjunctive queries (UCQs) studied in database theory, under bag semantics, may be viewed as a special case of the VCSPs that we consider. We obtain a complexity dichotomy for the case of acyclic UCQs and exemplarily use our methods to determine the complexity of a query that had remained open in the literature. Further, we conjecture that our hardness and tractability conditions match for resilience problems for UCQs.

RELATED WORK
1) The Complexity of Resilience Problems via Valued Constraint Satisfaction Problems.
https://arxiv.org/abs/2309.15654

2) The complexity of temporal constraint satisfaction problems. JACM 2010.
https://dl.acm.org/doi/10.1145/1667053.1667058

3) Complexity of Infinite-Domain Constraint Satisfaction. 2021.
https://www.cambridge.org/de/universitypress/subjects/mathematics/logic-categories-and-sets/complexity-infinite-domain-constraint-satisfaction?format=HB&isbn=9781107042841

4) Datalog and constraint satisfaction with infinite templates. JCSS 2013.
https://www.sciencedirect.com/science/article/pii/S0022000012001213

SHORT BIO
https://tu-dresden.de/mn/math/algebra/bodirsky
Since 3.2023: PI of the ERC Synergy Grant "POCOCOP" (jointly with Libor Barto, Charles University Prague, and Michael Pinsker, TU Wien)
Consolidator Grant of the ERC for the project "CSP-Infinity'' (2016-2021).
Since August 2014: Professor for Algebra and Discrete Structures at TU Dresden.
Starting Grant of the ERC for the project "Constraint Satisfaction Problems: Algorithms and Complexity'' (2011).
2008-2014: Chargé de Recherche of the CNRS at Ecole Polytechnique.
2005-2007: Postdoc in the DFG research group "Algorithms, Structure, Randomness'' and researcher at the Institute of Algorithms and Complexity at HU Berlin.
2004: PhD at Humboldt-University Berlin (supervision: Prof. Dr. Hans Jürgen Prömel) with a scholarship of the DFG research training group "Combinatorics, Geometry, and Computation''.
2001: Diploma in Computer Science (minor: mathematics) at Saarland University.

Title: 

Towards a Holistic Framework for Data-centric Responsible AI

Abstract: 

Responsible AI is becoming critical as AI is widely used in our everyday lives. Many companies that deploy AI publicly state that when training a model, we not only need to improve its accuracy, but also need to guarantee that the model does not discriminate against users (fairness), is resilient to noisy or poisoned data (robustness), protects personal information (privacy), and is interpretable (explainability) among others. In addition, these objectives are not only relevant to model training, but to all steps of machine learning starting from the data. A holy grail challenge is to support all of these objectives in one holistic framework that is also easy to use. To this end, I will give an overview of the data-centric responsible AI research in the Data Intelligence Lab at KAIST. I will first introduce fairness techniques throughout the machine learning process: fair model training using sample selection (ICLR’21), fair data labeling (VLDB’24, ACM SIGMOD’23), and fair data acquisition (ACM SIGMOD’21). Next, I will explain our works on other responsible AI objectives: fair and robust training (NeurIPS’21, ICML’20), privacy using disinformation (AAAI’23), and explainability-first clustering (AAAI’23). We believe we are just scratching the surface of a truly holistic solution and that many exciting challenges lie ahead.

Related papers:

• Roh et al., "FairBatch: Batch Selection for Model Fairness", ICLR 2021
• Zhang et al., "iFlipper: Label Flipping for Individual Fairness", SIGMOD 2023
• Tae et al., “Falcon: Fair Active Learning using Multi-armed Bandits”, VLDB 2024
• Tae and Whang, "Slice Tuner: A Selective Data Acquisition Framework for Accurate and Fair Machine Learning Models", SIGMOD 2021
• Roh et al., "Improving Fair Training under Correlation Shifts", ICML 2023
• Roh et al., "Sample Selection for Fair and Robust Training", NeurIPS 2021
• Heo and Whang, "Redactor: A Data-Centric and Individualized Defense against Inference Attacks", AAAI 2023
• Hwang and Whang, "XClusters: Explainability-first Clustering", AAAI 2023

Bio:

Steven Euijong Whang is an associate professor with tenure at KAIST EE and AI and leads the Data Intelligence Lab. His research interests include Responsible AI and Data-centric AI. He is an Associate Editor of IEEE TKDE (2023-2025), IEEE DE Bulletin (2023, 2024), and VLDB 2025. Previously he was a Research Scientist at Google Research and co-developed the data infrastructure of the TensorFlow Extended (TFX) machine learning platform. Steven received his Ph.D. in computer science in 2012 from Stanford University and his B.S. in computer science from KAIST in 2003. He was a Kwon Oh-Hyun Endowed Chair Professor (2020-2023) and received a Google AI Focused Research Award (2018, the first in Asia). Homepage: https://stevenwhang.com

Title: Explainable Disparity Compensation for Fair Ranking

Abstract: Ranking functions are commonly used to assist in decision-making in a wide variety of applications. Oftentimes, these ranking functions produce disparate results for different populations because of bias in the underlying data. In this talk, I will discuss transparent participation metrics to clarify the ranking process, by assessing the contribution of each parameter used in the ranking function in the creation of the final ranked outcome, using information about the ranking functions themselves, as well as observations of the underlying distributions of the parameter values involved in the ranking. In addition, I will propose easily explainable data-driven compensatory measures for ranking functions that rely on the generation of bonus points given to members of underrepresented groups to address disparity in the ranking function. The bonus points can be set in advance, and can be combined, allowing for considering the intersections of representations and giving better transparency to stakeholders. I will present efficient sampling-based algorithms to calculate the number of bonus points to minimize disparity, and present the results of an evaluation of the proposed metrics over a real-world school admissions dataset.

Related papers: 

• Abraham Gale, Amélie Marian: Explaining Ranking Functions. Proc. VLDB Endow. 14(4): 640-652 (2020)
• Abraham Gale, Amélie Marian: Explainable Disparity Compensation for Efficient Fair Ranking. ICDE 2024

Bio: Amélie Marian is an Associate Professor in the Computer Science Department at Rutgers University. Her research interests are in Explainable Rankings, Accountability of Decision-making Systems, Transparent Democracy, Personal Data, and Data Integration. Her recent public scholarship work on explaining the NYC School Admission lottery process to families, in collaboration with elected parent representatives, was instrumental in increasing transparency and accountability in the NYC high school application system. Amélie received her Ph.D. in Computer Science from Columbia University in 2005. She is the recipient of a Microsoft Live Labs Award, three Google Research Awards, and an NSF CAREER award.

Abstract:
Machine learning has been widely adopted in real-world applications that have great social impact, and fairness in automatic decision-making systems has become an arising concern. It has been shown that without proper intervention, machine learning models can exhibit biases against certain demographic groups. However, simply excluding sensitive attributes (such as race, gender, religion) during training does not improve fairness, given the distributional disparities across different demographic groups. Although various methods have been proposed to mitigate biases, many rely on access to predefined sensitive information during training, presenting challenges in practical settings where collecting such data may be infeasible due to legal or regulatory constraints. Additionally, for many automatic decision-making systems, it is important to ensure fairness across multiple sensitive attributes, which presents a challenge for conventional methods formulated with only one predefined sensitive attribute.

In response to these challenges, I will discuss our recent work focusing on fair machine learning without demographics. We provide rigorous theoretical guarantees of fairness for our approach. We also propose strategies to enhance efficiency, particularly in addressing fairness concerns in large-scale transformer models. Notably, our approach demonstrates effectiveness in improving fairness in large-scale transformers with substantially lower energy consumption.

Related papers:

• Junyi Chai, Taeuk Jang, Xiaoqian Wang: Fairness without Demographics through Knowledge Distillation. NeurIPS 2022
• Shenyu Lu, Yipei Wang, Xiaoqian Wang: Debiasing Attention Mechanism in Transformer without Demographics. ICLR 2024

Bio: Xiaoqian Wang is an Assistant Professor of Electrical and Computer Engineering at Purdue University. She received her Ph.D. degree from the University of Pittsburgh in 2019, and the B.S. degree from Zhejiang University in 2013. She focuses on designing novel machine learning models to improve interpretability, fairness, and robustness. She also work on the intersection of machine learning and bioinformatics, healthcare.  She has published more than 40 papers in top conferences and journals including NeurIPS, ICML, ICLR, AAAI, IJCAI, CVPR, KDD, ICDM, RECOMB, ECCB, IPMI, MICCAI, etc. She received an NSF CAREER award in 2022, and an AAAI distinguished paper award in 2023.

Title: Finding Best Tuple via Error-prone User Interaction

Abstract:
In the literature of the database community, there are a lot of studies about finding a utility function from a user (representing the user's preference), via interaction with the user by asking a number of questions each requiring him/her to compare 2 points for choosing a more preferred point, in order to find the best tuple in the database containing a lot of tuples. In the real world, the user may make mistakes (carelessly), which means that s/he may answer some of the questions wrongly. Unfortunately, existing interaction algorithms may find the undesirable point based on the wrongly learnt utility function because they assume that all answers from the user are 100% correct. In particular, even if the user answers only 1 wrong answer, the output of the existing algorithms may be far away from the users' real need. Motivated by this, in this paper, we propose a new problem of finding the most interesting point via interaction which is robust to possible mistakes made by a user. Besides, we propose (1) an algorithm that asks an asymptotically optimal number of questions when the dataset contains 2 dimensions and (2) two algorithms with provable performance guarantee when the dataset contains d dimensions where d = 2. Experiments on real and synthetic datasets show that our algorithms outperform the existing ones with a higher accuracy with only a small number of questions asked.

Related Paper:
Qixu Chen, Raymond Chi-Wing Wong: Finding Best Tuple via Error-prone User Interaction. ICDE 2023: 1803-1816. https://home.cse.ust.hk/~raywong/paper/icde23-interactionError.pdf

Biography:
Raymond Chi-Wing Wong is a Professor in Computer Science and Engineering (CSE) of The Hong Kong University of Science and Technology (HKUST). He is currently the associate head of Department of Computer Science and Engineering (CSE). He was the associate director of the Data Science & Technology (DSCT) program (from 2019 to 2021), the director of the Risk Management and Business Intelligence (RMBI) program (from 2017 to 2019), the director of the Computer Engineering (CPEG) program (from 2014 to 2016) and the associate director of the Computer Engineering (CPEG) program (from 2012 to 2014). He received the BSc, MPhil and PhD degrees in Computer Science and Engineering in the Chinese University of Hong Kong (CUHK) in 2002, 2004 and 2008, respectively. In 2004-2005, he worked as a research and development assistant under an R&D project funded by ITF and a local industrial company called Lifewood.
He received 40 awards. He published 119 conference papers (e.g., SIGMOD, SIGKDD, VLDB, ICDE and ICDM), 46 journal/chapter papers (e.g., TODS, DAMI, TKDE, VLDB journal and TKDD) and 1 book. He reviewed papers from conferences and journals related to data mining and database, including VLDB conference, SIGMOD, TODS, VLDB Journal, TKDE, TKDD, ICDE, SIGKDD, ICDM, DAMI, DaWaK, PAKDD, EDBT and IJDWM. He is a program committee member of conferences, including SIGMOD, VLDB, ICDE, KDD, ICDM and SDM, and a referee of journals, including TODS, VLDBJ, TKDE, TKDD, DAMI and KAIS.
His research interests include database, data mining and artificial intelligence.

Title: 
Optimization Opportunities in Human-in-the-loop Systems

Abstract: 
An emerging trend is to leverage an under-explored and richly heterogeneous pool of human knowledge inside machine algorithms, a practice popularly termed as human-in-the-loop (HIL) process. A wide variety of applications, starting from query processing to text translation, feature engineering, or even human decision making in complex uncertain environments stand to benefit from such synergistic man-machine collaboration. This talk will discuss ongoing federally funded research projects in my lab, some recent research results, and impacts that investigate a variety of optimization opportunities inside such HIL systems, considering the roles and responsibilities of three key stakeholders - humans (workers), machines (algorithms), and platforms (online infrastructure where the work takes place). I will highlight data management challenges and scalability opportunities present in these problems.

Related papers:
a. Md Mouinul Islam, Dong Wei, Baruch Schieber, Senjuti Basu Roy:
Satisfying Complex Top-k Fairness Constraints by Preference Substitutions. Proc. VLDB Endow. 16(2): 317-329 (2022)
b. Dong Wei, Md Mouinul Islam, Baruch Schieber, Senjuti Basu Roy:
Rank Aggregation with Proportionate Fairness. SIGMOD Conference 2022: 262-275
c. Sepideh Nikookar, Paras Sakharkar, Sathyanarayanan Somasunder, Senjuti Basu Roy, Adam Bienkowski, Matthew Macesker, Krishna R. Pattipati, David Sidoti:
Cooperative Route Planning Framework for Multiple Distributed Assets in Maritime Applications. SIGMOD Conference 2022: 1518-1527

Speaker Bio:
Senjuti Basu Roy is the Panasonic Chair in Sustainability and an Associate Professor in the Department of Computer Science at the New Jersey Institute of Technology. Her research focus lies at the intersection of data management, data exploration, and AI, especially enabling human-machine analytics in scale. Senjuti has published more than 85 research papers in high impact data management and data mining conferences and journals. She is currently serving as the track chair of Economics, Online Markets, and Human Computation track of The ACM Web Conference 2024, an Associate Editor of ICDE 2024,  SIGMOD 2025, VLDB Journal, and Information Systems. She is a recipient of the NSF CAREER Award, a Presidential Early Career Award for Scientists and Engineers (PECASE) nominee, and one of the 100 invited early career engineers selected by the National Academy of Engineering in 2021.

TITLE
A survey of ideal matrices

ABSTRACT
A 0,1 matrix M is ideal if the set covering polyhedron {x>=0: Mx>=1} is integral. This notion remains a cornerstone in the theory of integer and linear programming. In this survey talk, I will talk about the historical context of ideal matrices and related objects, present some old and recent findings, and end with some exciting conjectures in the area.

RELATED WORK
* Abdi and Lee: Deltas, extended odd holes and their blockers. Journal of Combinatorial Theory, Series B 136:193-203 (2019), https://doi.org/10.1016/j.jctb.2018.10.006
* Abdi, Cornuejols, Lee: Intersecting restrictions in clutters. Combinatorica 40:605-623 (2020), https://doi.org/10.1007/s00493-020-4076-2
* Abdi, Cornuejols, Guenin, Tuncel: Testing idealness in the filter oracle model. Operation Research Letters 50(6): 753-755 (2022), https://doi.org/10.1016/j.orl.2022.11.004

SHORT BIO
I obtained my PhD degree in Mathematics at the University of Waterloo in 2018; the title of my dissertation was "Ideal Clutters". I then spent a year and a bit as a postdoctoral fellow at Carnegie Mellon University, before joining the London School of Economics as a tenure-track assistant professor in 2019. My research expertise are broadly in Combinatorial Optimization, Integer and Linear Programming, Matroid Theory, and Graph Theory. More specifically, the focus of my research so far has been the study of Ideal Clutters, a polyhedrally defined class of objects with deep connections to the areas mentioned, and its applications.
https://www.ahmadabdi.com/

Title: Beyond Correlation: Harnessing Causal Inference for Enhanced Data Insights

Causal reasoning has emerged as a crucial element in artificial intelligence, healthcare, and economics, offering a powerful tool for understanding the complex relationships between variables. However, the practical application of causal reasoning methods has been hindered by the extensive and challenging data preparation required. This talk will explore the promise of leveraging causal reasoning methods for various data analytics tasks. We will introduce Hyper, a framework that answers what-if and how-to queries over relational data, which are commonly required for risk management, policy design, exploratory data analysis, and experiment design. By extending SQL syntax, users can express hypothetical queries. We have developed novel techniques that capture the probabilistic dependencies between attributes in the form of a causal graph and use it to efficiently answer user queries. Furthermore, we will discuss the application of causal reasoning techniques for other data preparation and analytics tasks such as fairness, explainability, and debugging. The talk will conclude with a glimpse of how data management techniques can be invaluable for causal discovery, highlighting the need for further exploration of this synergy.

Related papers:
https://dl.acm.org/doi/abs/10.1145/3514221.3526149
https://dl.acm.org/doi/abs/10.1145/3514221.3517864

Bio: Sainyam Galhotra is an Assistant Professor in the Computer Science department at Cornell University. Prior to this role, he held the position of Computing Innovation Postdoctoral Fellow at the University of Chicago. His research is driven by the ambition to establish the foundational principles of responsible data science, which is pivotal for the efficient development and deployment of reliable data analytics applications. Galhotra's work is distinguished by its interdisciplinary approach, integrating techniques from Data Management, Probabilistic Methods, Causal Inference, Machine Learning, and Software Engineering. He earned his Ph.D. from the University of Massachusetts Amherst. His contributions to the field have been recognized with several accolades, including the Best Paper Award at FSE 2017, and the Most Reproducible Paper Award at SIGMOD in both 2017 and 2018. Additionally, he is a DAAD AInet Fellow and the inaugural recipient of the Krithi Ramamritham Award at UMass, an honor bestowed for outstanding contributions to database research.

Title: Data-driven Solutions for AI Responsibility

Abstract: Data-driven tasks are used in assisting decisions in many domains that greatly affect people’s lives. Though these tasks are efficient, they inevitably inherit bias in the input data and show problematic behaviors. Deploying unfair algorithms in critical domains will reinforce and legitimize the inequitable treatment of groups who are already disadvantaged. Moreover, the deployment of these tasks lacks transparency and accountability. In this talk, I will introduce my research on mitigating bias in two types of data-driven tasks: rankings and classifications, which are frequently set up to automate decision-making processes. I will also highlight some ongoing work of promoting responsible AI by data-driven solutions.

Related papers:
1. https://arxiv.org/abs/2303.17566 (ICDE 24 preprint)
2. https://drops.dagstuhl.de/storage/00lipics/lipics-vol192-forc2021/LIPIcs.FORC.2021.7/LIPIcs.FORC.2021.7.pdf
3. https://dl.acm.org/doi/pdf/10.1145/3555041.3589405

Bio: Dr. Ke Yang is an Assistant Professor in Computer Science in the College of Sciences, at the University of Texas San Antonio. She found the Cohort for AI REsponsibility (CAREAI) at UTSA. Her research centers around data management and machine learning, emphasizing algorithmic fairness, model explanation, and the social impact of AI and machine learning. Prior to joining UTSA, she was a Postdoctoral Research Associate in the College of Information and Computer Sciences, at the University of Massachusetts Amherst.  Dr. Yang received her Ph.D. degree from New York University. She has received a Postdoctoral Fellowship from the Center for Data Science at UMass and a Pearl Brownstein Doctoral Research Award from the Tandon School of Engineering at NYU.

Title: Modeling File Structure for Data Preparation
Speaker:  Gerardo Vitagliano 
Time:  11am ET
Location:  WVH 446, 440 Huntington Ave., Boston
Host:  Renée Miller 

Tabular data is at the core of many data-driven processes. Tabular files are referred to as "structured" data, implying that they can be easily processed automatically. However, in real-world processes, extensive data preparation is often required to parse, repair, or transform these files before leveraging their contents.  We overview several models to represent the heterogeneous structure of real-world files: a formal model, an implicit model, and an embedding model.

We present MaGRiTTE, a transformer-based architecture that models file structure as vectorial embeddings, and apply it to solve automated data preparation tasks.Bio:

Bio:  Gerardo Vitagliano is a PhD student at the Hasso Plattner Institute, in the Information System group of Prof. Felix Naumann. His research interests are structural data preparation, automated data management, and the intersection between machine learning and data quality. 

Papers:  http://www.vldb.org/pvldb/vol15/p646-vitagliano.pdf
https://openproceedings.org/2021/conf/edbt/p32.pdf

Speaker:  Michael Cafarella, MIT

Location:  WVH 446 
Host:  Renée Miller (contact for zoom link)

Abstract:
Causal reasoning is a growing part of AI and has become a core part of economists’ toolkit. But applying causal reasoning methods in practice has required a large amount of careful, difficult database preparation. As a result, causal methods have not penetrated into more mundane domains that could benefit from causal discovery.
This talk will describe early work on how to apply causal methods to the practical domain of distributed system debugging. It offers a framework for processing log files and automatically constructing a representation that is amenable to causal interpretation. We also present algorithms for how to efficiently use this representation for causal discovery of dynamics of a distributed system, and for marshaling scarce user curation attention for building domain knowledge needed to answer practical systems questions.
Title: Sawmill: Applying Causal Reasoning to System Logs

Bio:  Michael Cafarella is a Principal Research Scientist in the MIT Computer Science and Artificial Intelligence Lab. His research interests include databases, information extraction, data integration, and applying data-intensive methods to economics. He has published extensively in venues such as SIGMOD, VLDB, and elsewhere. Mike received his PhD from the University of Washington, Seattle, in 2009. His academic awards include the NSF CAREER award, the Sloan Research Fellowship, and the VLDB Ten-Year Best Paper award. In addition to his academic work, he co-founded the widely-used Hadoop open-source project and information extraction startup Lattice Data.√

Abolfazl Asudeh, UIC

Maximizing Neutrality in News Ordering

noon ET Nov 8,  WVH 446, 440 Huntington Ave, Boston

Host:  Renee Miller (contact for zoom link)

The detection of fake news has received increasing attention over the past few years, but there are more subtle ways of deceiving one's audience. In addition to the content of news stories, their presentation can also be made misleading or biased. In this work, we study the impact of the ordering of news stories on audience perception. We introduce the problems of detecting cherry-picked news orderings and maximizing neutrality in news orderings. We prove hardness results and present several algorithms for approximately solving these problems. Furthermore, we provide extensive experimental results and present evidence of potential cherry-picking in the real world.

Paper: KDD 2023

Bio:  Abolfazl Asudeh is an assistant professor of Computer Science at the University of Illinois Chicago and the director of Innovative Data Exploration Laboratory (InDeX Lab). He is a VLDB Ambassador, and a regular PC member of Database flagship venues.His research spans to different aspects of Big Data and Data Science, including Data Management, Information Retrieval, and Data Mining, for which he designs efficient, accurate, and scalable algorithmic solutions. Data Equity Systems, Algorithmic Fairness, and Data-centric AI are his major focus in research. His research interest also includes: Ranking, Social Networks Analysis, Large-Scale Computation on Limited Resources, Computational Fact Checking, Data management for Machine Learning, Web Databases, and applied Computational Geometry.

Speaker: Boris Glavic, IIT-Chicago

Title:  UNCERTAIN DATA MANAGEMENT FOR RESPONSIBLE DATA                            MANAGEMENT      
     Lightweight Uncertain Data Models to The Rescue

Host:  Renee Miller (contact for zoom link)

noon ET Nov 10, WVH 446, 440 Huntington Ave, Boston

Uncertainty arises naturally in many application domains due to data
entry errors, sensor errors and noise, uncertainty in information
extraction and parsing, selection bias, ambiguity from data integration,
and heuristic data preparation & cleaning techniques. Analyzing
uncertain data without accounting for its uncertainty can create hard to
trace errors with severe real world implications. Thus, tracking
uncertainty in data is essential for responsible data management. In the
first part of this talk, I will discuss how uncertainty in data
collection and preparation leads to biased and untrustworthy analysis
results. Afterwards, I will introduce the audience to methods for
modeling uncertainty and for determining how uncertainty impacts
analysis results. Specifically, I will focus on uncertainty-annotated
databases (UA-DBs), a light-weight and practical uncertain data model.
UA-DBs accommodate a wide range of uncertain data sources, can
approximate many existing incomplete and probabilistic data models as
well as can approximate the uncertainty arising naturally in many data
cleaning techniques, and, most importantly, query answering over UA-DBs
is efficient for a very large class of queries.

Uncertainty Annotated Databases - a Lightweight Approach for Approximating Certain Answers. Su Feng, Aaron Huber, Boris Glavic, Oliver Kennedy.  SIGMOD19

Efficient Uncertainty Tracking for Complex Queries with Attribute-Level Bounds. Su Feng, Aaron Huber, Boris Glavic, Oliver Kennedy.  SIGMOD21

Towards Diversified and Privatized Data Preparation 

Speaker:  Fei Chiang, McMaster University

Host:  Renee Miller (contact for zoom link)

Recent privacy legislation has aimed to restrict and control the amount of personal data published by companies and shared with third parties. Much of this real data is not only sensitive requiring anonymization but also contains characteristic details from a variety of individuals. This diversity is desirable in many applications ranging from Web search to drug and product development. Unfortunately, data anonymization techniques have largely ignored diversity in its published result. This inadvertently propagates underlying bias in subsequent data analysis. We study the problem of finding a diverse anonymized data instance where diversity is measured via a set of diversity constraints.   In this talk, I will present some of our work on finding a diverse anonymized instance, as well as extensions to data publishing techniques to consider data semantics while protecting sensitive values. 

Bio: 

Fei Chiang is an Associate Professor in the Department of Computing and Software (Faculty of Engineering), the Director of the Data Science Research Group, and a Faculty Fellow at the IBM Centre for Advanced Studies. She served as an inaugural Associate Director of the MacData Institute. Her research interests and industrial experience is in data management, spanning data quality, data profiling, temporal graphs, and database systems. She holds four patents for her work in self-managing database systems.  She received a 2018 Ontario Early Researcher Award, and a 2023 VLDB Distinguished Reviewer Award.  She received her M. Math from the University of Waterloo, and B.Sc and PhD degrees from the University of Toronto, all in Computer Science.

TITLE
Viper: A Fast Snapshot Isolation Checker

ABSTRACT
Snapshot isolation (SI) is supported by most commercial databases and is widely used by applications. However, checking SI today—given a set of transactions, checking if they obey SI—is either slow or gives up soundness.

We present Viper, an SI checker that is sound, complete, and fast. Viper checks black-box databases and hence is transparent to both users and databases. To be fast, viper introduces BC-polygraphs, a new representation of transaction dependencies. A BC-polygraph is acyclic iff transactions are SI, a theorem that we prove. Viper also introduces heuristic pruning, an optimization to accelerate checking SI by leveraging common knowledge of real-world database implementations. Besides vanilla SI, viper supports major SI variants including Strong SI, Generalized SI, and Strong Session SI. Our experiments show that given the same time budget, viper improves over baselines by 15× in the workload sizes being checked.

RELATED PAPERS
* Zhang, Ji, Mu, Tan. Viper: A Fast Snapshot Isolation Checker. (https://www.khoury.northeastern.edu/home/zhangjian3/attaches/eurosys23-final117.pdf)
* Adya. Weak Consistency: A Generalized Theory and Optimistic Implementations for Distributed Transactions (https://pmg.csail.mit.edu/papers/adya-phd.pdf
* Crooks, Pu, Alvisi, Clement. Seeing is Believing: A Client-Centric Specification of Database Isolation (https://www.cs.cornell.edu/lorenzo/papers/Crooks17Seeing.pdf")

BIO
Jian Zhang is a second-year PhD student of Khoury College of Computer Sciences at Northeastern University, advised by Cheng Tan. His main research interests are verifying correctness of distributed systems and building efficient distributed systems. His recent projects mainly focus on verifying isolation levels of databases.

TITLE
Optimal routings and crowdsourcing (and some "work in progress")

ABSTRACT
In this seminar I will present three problems involving crowdsourcing and routing and whose solution involve the use of skylines. The main theme among those problems is to present so-called "gig-workers" with optimal alternative routes which optimize two conflicting criteria, e.g., minimizing the cost of a path while maximizing rewards from completing tasks in such a path. I will also introduce two problems that I am currently working on: leveraging large language models for spatially oriented tasks such as trajectory imputation, and how to detect and fix data that causes previously learned models to fail.

READINGS:
- C.F. Costa and M.A. Nascimento, Last Mile Delivery Considering Time-Dependent Locations, ACM SIGSPATIAL, 2021. https://dl.acm.org/doi/10.1145/3474717.3483919
- C.F. Costa and M.A. Nascimento, In-Route Task Selection in Spatial Crowdsourcing, ACM Trans. Spatial Algorithms and Systems, 2020. https://dl.acm.org/doi/10.1145/3368268
- S. Anwar, F. Lettich, M.A. Nascimento, Towards A Personal Shopper's Dilemma: Time vs Cost. ACM SIGSPATIAL/GIS 2020. https://dl.acm.org/doi/10.1145/3397536.3422276

BIO
Mario Nascimento is currently with the Northeastern University’s Khoury College, Vancouver Campus, serving as its first Director of Pacific Northwest Research. His main research interests are in the area of spatiotemporal data management and, according to Google Scholar, my publications have been cited 4,600+ times, yieding an H-index of 33. He is also serving as the PC co-Chair for ACM SIGSPATIAL 2023, and General co-Chair for SSTD 2023. Among many other service-oriented roles, Mario has also served as ACM SIGMOD‘s Information Director, ACM SIGMOD Record‘s Editor and VLDB Journal‘s Editorial Board member.

TITLE
Associative Array Algebra of Big Data, Graphs, & Machine Learning

ABSTRACT
Rectangular arrays abound in mathematics and computer science, e.g., spreadsheets, databases, matrices, and graphs (the last by way of incidence or adjacency arrays). Despite none of these instantiations being new, the ever-increasing amount of data means that novel approaches are required for storing, operating on, and interpreting these huge sets of data. Some such approaches include leveraging semirings and sparse linear algebra, as in GraphBLAS and the Dynamic Distributed Dimensional Data Model (D4M). Associative arrays exemplify the marrying of semirings and sparse linear algebra, providing a common interface for the aforementioned varying types of rectangular arrays--generalizing them yet still enforcing familiar and important mathematical properties. The main features of associative arrays that differ from traditional treatments of, say, matrices are the fact that row and column indices may be taken from arbitrary sets (as opposed to finite initial segments of the positive or nonnegative integers) and entries are taken in semirings (as opposed to fields or rings).

In this seminar I will explore the mathematical theory of semirings and associative array algebra (including both positive and negative results), their use and implementations in GraphBLAS and D4M, and current and recent applications.

READINGS
* Associative Arrays: Unified Mathematics for Spreadsheets, Databases, Matrices, and Graphs (https://arxiv.org/abs/1501.05709)
* Python implementation of the Dynamic Distributed Dimensional Data Model (https://arxiv.org/abs/2209.00602)
* GraphBLAS Mathematics (https://www.mit.edu/~kepner/GraphBLAS/GraphBLAS-Math-release.pdf)

BIO
Dr Hayden Jananthan is a mathematician whose focus is on applying traditionally "pure" mathematics tools to applied contexts, particularly on the theory of associative arrays as a tool to address big data and machine learning problems. More recently, Dr Jananthan has been applying the theory of associative arrays and the study of heavy-tailed statistics to better understand the foundational characteristics of malicious network traffic. In addition to his research, Dr Jananthan is passionate about making pure mathematics more accessible to STEM professionals in general. Dr Jananthan received his PhD from Vanderbilt University, where his doctoral research was focused on computability theory. As recognition for his research and teaching contributions, he received the Bjarni Jonsson Prize for Research and the B. F. Bryant Award for Excellence in Teaching.

TITLE
Direct Model Editing

ABSTRACT
Can we understand large deep networks well enough to reprogram them by changing their parameters directly? In this talk I will talk about Direct Model Editing, a framework for understanding and modifying the structure of their internal computations using causal interventions. We will consider examples in computer vision and NLP: I will talk about how to probe and rewrite computations within a StyleGAN to alter compositional rules that govern rendering of realistic images, and I will describe how to edit specific factual knowledge within a large language model, directly tracing and modifying parameters of GPT.

RELATED WORK
* Locating and Editing Factual Associations in GPT. https://rome.baulab.info/
* Mass Editing Memory in a Transformer. Preprint. https://memit.baulab.info/
* GAN Dissection. PNAS 2020. https://gandissect.csail.mit.edu/
* StyleSpace Analysis: Disentangled Controls for StyleGAN Image Generation. CVPR 2021. https://github.com/betterze/StyleSpace

BIO
David Bau is Assistant Professor at the Northeastern University Khoury College of Computer Science. He received his PhD from MIT and AB from Harvard, and he has previously worked at Google and Microsoft. He is known for his network dissection studies of individual neurons in deep networks and has published research on the interpretable structure of learned computations in large models in PNAS, CVPR, NeurIPS, and SIGGRAPH. Prof. Bau is also coauthor of the textbook, Numerical Linear Algebra.

TITLE
Ontologies for Data Access and Data Quality

ABSTRACT
Ontology-mediated query answering (OMQA) is a promising approach to data access and integration that has been extensively studied by the AI and database communities. Ontologies, which are typically specified using logic-based formalisms, serve both to provide a user-friendly vocabulary for query formulation and to capture domain knowledge, which can be exploited by reasoning algorithms to infer implicit information and return more complete query results. While the addition of an ontology brings significant benefits, it also renders the query answering task more computationally involved, which has led both to theoretical studies aimed at understanding the complexity of OMQA for a range of ontology and query languages, as well as the development of new algorithmic techniques.

I will start this talk by giving a brief introduction to OMQA, outlining the two main algorithmic approaches (query rewriting and materialization), and highlighting some notable research questions and results related to query rewriting. In the second part of the talk, I will discuss data quality in the context of OMQA, in particular, how to handle data inconsistencies.

RELATED PAPERS
1. Bienvenu : A Short Survey on Inconsistency Handling in Ontology-Mediated Query Answering. Special issue on "Ontologies and Data Management" of Künstliche Intelligenz 34(4): 443-451. 2020.
https://www.labri.fr/perso/meghyn/papers/Bienvenu-KI20.pdf

2. Bienvenu, Kikot, Kontchakov, Podolskiii, Zakharyaschev: Ontology-Mediated Queries: Combined Complexity and Succinctness of Rewritings via Circuit Complexity. Journal of the ACM (JACM) 65:5, 2018.
https://www.labri.fr/perso/meghyn/papers/BieKikKonPodZak-JACM18.pdf

3. Bienvenu, ten Cate, Lutz, Wolter: Ontology-based Data Access: A Study Through Disjunctive Datalog, CSP, and MMSNP. ACM Transactions on Database Systems (TODS) 39(4): 33:1 -33:44.
https://www.labri.fr/perso/meghyn/papers/BieCatLutWol-TODS14.pdf

BIO
Meghyn Bienvenu is a CNRS research director at LaBRI (Bordeaux, France). Born in Canada, she obtained her undergraduate degree from the University of Toronto before moving to France to continue her studies at the University of Toulouse. Her PhD thesis, defended in 2009, was awarded the AFIA Prize for best French dissertation in artificial intelligence. Her research interests span a range of topics in knowledge representation and reasoning and database theory, with a main focus on description logic ontologies and their use in querying data. She currently leads an ANR AI Chair on the topic of intelligent handling of imperfect data. Since January 2023, she is co-director of the CNRS National Research Network RADIA (Reasoning, Learning, and Decision in AI). She is an associate editor of Artificial Intelligence Journal (AIJ) and ACM Transactions on Computational Logic (TOCL), and she served as PC co-chair for KR 2021, the leading conference on knowledge representation and reasoning. Her research has been recognized by an invited Early Career Spotlight talk at IJCAI’16 and the 2016 CNRS Bronze Medal in the area of computer science.

TITLE
Building Natural Language Interfaces for Database with and without AI

ABSTRACT
Information systems are the core of modern enterprises and scientific exploration. They are often based on fundamental research developed in the database community. While enterprise data is typically stored in relational databases, data-intensive scientific disciplines such as bioinformatics often store their data in graph databases. To query these databases, end-users need to know the formal query languages SQL or SPARQL as well as the logical structure of the databases. However, even for technology experts it is very challenging to write the right queries to retrieve the desired data. Hence, a large part of the end-users is basically not able to effectively query their databases.

In this talk we discuss how to build intelligent information systems that enable end-users to talk to their data similar to humans. The major goal is to combine artificial intelligence with human intelligence for novel ways of data exploration. In particular, we will show how we have built various natural language interfaces for databases using pattern-based and machine learning-based approaches to significantly increase the productivity of scientists and knowledge workers when interacting with data. We demonstrate that our system INODE (Intelligent Open Data Exploration), which we have been building as part of a European Union project with 9 partners across Europe, is uniquely accessible to a wide range of users from large scientific communities to the public. Finally, we elaborate on the lessons learned when developing such a system and discuss how the technology can be enhanced by researchers or knowledge workers for exploring their own databases in natural language.

MAJOR REFERENCES
* Amer-Yahia, Koutrika, Braschler, Calvanese, Lanti, Lücke-Tieke, ... & Stockinger. INODE: Building an End-to-End Data Exploration System in Practice. ACM SIGMOD Record. 2022.
https://sigmodrecord.org/?smd_process_download=1&download_id=%2012688
* Sima, Mendes de Farias, Anisimova, Dessimoz, Robinson-Rechavi, Zbinden, Stockinger. Bio-SODA UX: enabling natural language question answering over knowledge graphs with user disambiguation. Distributed and Parallel Databases. 2022.
https://link.springer.com/article/10.1007/s10619-022-07414-w
* Brunner, Stockinger. ValueNet: A Natural Language-to-SQL System that Learns from Database Information. ICDE 2021.
https://arxiv.org/pdf/2006.00888

BIO
Prof. Dr. Kurt Stockinger is Professor of Computer Science, Director of Studies in Data Science at Zurich University of Applied Sciences (ZHAW) and Co-Head of the ZHAW Datalab in Switzerland. He is also an external lecturer at University of Zurich. Kurt Stockingers’ research focuses on Data Science with emphasis on Big Data, Natural Language Query Processing, Query Optimization and Quantum Computing. Essentially, his research interests are at the intersection of databases, natural language processing and machine learning. He is also on the Advisory Board of Callista Group AG. Previously Kurt Stockinger worked at Credit Suisse in Zurich, Switzerland, at Lawrence Berkeley National Laboratory in Berkeley, California, at California Institute of Technology, California as well as at CERN in Geneva, Switzerland. He holds a Ph.D. in computer science from CERN / University of Vienna.

TITLE
Applications of Information Inequalities to Database Theory Problems

ABSTRACT
The talk describes applications of information inequalities to three database problems: upper bounds of a query's output (a.k.a. pessimistic cardinality estimation), the query domination problem (a.k.a. query containment under bag semantics), and the implication problem for approximate integrity constraints. No prior knowledge of query processing or information theory is needed: all required concepts and results from information inequalities are introduced here, gradually, and motivated by database problems.

RELATED PAPERS
* Applications of Information Inequalities to Database Theory Problems, 2023. (invited talk at LICS'2023)
https://arxiv.org/abs/2304.11996
* Kyle B. Deeds, Dan Suciu, Magdalena Balazinska: SafeBound: A Practical System for Generating Cardinality Bounds. Proc. ACM Manag. Data 1(1): 53:1-53:26 (2023)
https://dl.acm.org/doi/pdf/10.1145/3588907
* Mahmoud Abo Khamis, Phokion G. Kolaitis, Hung Q. Ngo, Dan Suciu: Bag Query Containment and Information Theory. ACM Trans. Database Syst. 46(3): 12:1-12:39 (2021)
https://dl.acm.org/doi/abs/10.1145/3472391
* Batya Kenig, Dan Suciu: Integrity Constraints Revisited: From Exact to Approximate Implication. Log. Methods Comput. Sci. 18(1) (2022)
https://drops.dagstuhl.de/opus/volltexte/2020/11942/pdf/LIPIcs-ICDT-2020-18.pdf
* Batya Kenig, Pranay Mundra, Guna Prasaad, Babak Salimi, Dan Suciu: Mining Approximate Acyclic Schemes from Relations. SIGMOD Conference 2020: 297-312
https://dl.acm.org/doi/pdf/10.1145/3318464.3380573

BIO
Dan Suciu is a Microsoft Endowed Professor in the Paul G. Allen School of Computer Science and Engineering at the University of Washington. Suciu is conducting research in data management, on topics such as query optimization, probabilistic data, data pricing, parallel data processing, data security. He is a co-author of two books Data on the Web: from Relations to Semistructured Data and XML, 1999, and Probabilistic Databases, 2011. He received the ACM SIGMOD Codd Innovation Award, received several best paper awards and test of time awards, and is a Fellow of the ACM. Suciu is currently an associate editor for the Journal of the ACM. Suciu's PhD students Gerome Miklau, Christopher Re and Paris Koutris received the ACM SIGMOD Best Dissertation Award in 2006, 2010, and 2016 respectively, and Nilesh Dalvi was a runner up in 2008.

TITLE
Relational Programming

ABSTRACT
Modern data analytics demands capabilities beyond those offered by traditional data management systems. On one hand, today’s data come in many different forms, be it graphs, matrices, or tensors; on the other, analysis tasks are becoming more and more complex, often involving complex operations running iteratively. These new demands call for new ways of working with data and innovative data analytic systems. In this talk, I make the case for relational programming, an emerging programming paradigm for modern data analytics. I describe a new programming language, a new algorithm for the relational join, and optimization techniques for relational programs. I conclude with my vision for a relational future of programming.

RELATED PAPERS
1) Optimizing Recursive Queries with Program Synthesis (https://arxiv.org/abs/2202.10390)
2) Free Join: Unifying Worst-Cast Optimal and Traditional Joins (in submission)
3) SPORES: Sum-Product Optimization via Relational Equality Saturation for Large Scale Linear Algebra (http://www.vldb.org/pvldb/vol13/p1919-wang.pdf)

BIO
Yisu Remy Wang is a PhD student at University of Washington. He works at the intersection of Databases and Programming Languages. He develops languages, theory and tools to help data analysts conduct complex tasks efficiently. His research has been recognized with best paper awards from PODS, POPL and OOPSLA.

TITLE
On the Complexity of Joins on Interval Data

ABSTRACT 
Interval data abound in spatio-temporal databases, as they represent spatial objects or time periods. This work established the complexity of Boolean Conjunctive Queries (IJs) with Intersection Joins, which are joins often applied to spatio-temporal databases. It was found that IJs can be solved by known algorithms for Boolean Conjunctive Queries. This can be accomplished by reducing the former problem to the latter. Additionally, this work showed that a reverse reduction could be used to derive lower bounds as well. Last but not least, a new acyclicity class emerged through this work to characterise syntactically the IJs that can be computed in linear time (up to logarithmic factors).
 
RELATED PAPERS
This talk is based on the article with title "The Complexity of Boolean Conjunctive Queries with Intersection Joins" which was published in PODS 2022.(https://dl.acm.org/doi/abs/10.1145/3517804.3524156)
 
BIO 
Antonia Kormpa has studied computer engineering and informatics at the University of Patras, Greece. She is currently pursuing a PhD in computer science at the University of Oxford, UK. Her research is about the complexity of join queries on interval data.

TITLE
Machine Learning Enhanced Query Scheduling and Execution Operations in Database Systems

ABSTRACT
Database systems have traditionally relied on empirical approaches and handcrafted rules that encode human intuitions or heuristics to store large-scale data and process user queries over them. These well-tuned empirical approaches and rules work well for the general-purpose case, but are seldom optimal for any actual application because they are not tailored for the specific application properties (e.g., user workload patterns). Further, they fail to consider the complex interaction with the environment running the database systems (e.g., hardware and operating system). In this talk, I will show how we leverage machine learning to instance-optimize the performance of the query execution engine, the heart of any database system, using efficient learned models, data structures, and algorithms for three essential operations: query scheduling, hashing, and join processing. 
 
RELATED PAPERS
1. Learned query scheduling: https://dl.acm.org/doi/abs/10.1145/3514221.3526158
2. Learned hashing: https://dl.acm.org/doi/10.14778/3570690.3570702
3. Learned in-memory joins: https://arxiv.org/abs/2111.08824  

BIO
Ibrahim Sabek is a postdoctoral associate at the MIT Data Systems Group and an NSF/CRA Computing Innovation Fellow, working with Tim Kraska and Michael Cafarella. Before that, he completed his Ph.D. at the University of Minnesota, Twin Cities, under the supervision of Mohamed Mokbel. Ibrahim is interested in building the next generation of machine learning-empowered data management, processing, and analysis systems. His research focuses on deeply understanding machine learning and systems techniques, resulting in entirely new designs, algorithms, and data structures for data-intensive systems and applications. His research areas broadly include instance-optimized database systems, scalable knowledge base construction, big spatial data management and analysis, and data processing in containerization environments. For more information about Ibrahim, please check his website: http://people.csail.mit.edu/ibrahimsabek/

TITLE
Acyclic Schemas and Information Theory
 
ABSTRACT
Acyclic schemas possess known benefits for database design, speeding up queries, and reducing space requirements. An acyclic join dependency (AJD) is lossless with respect to a universal relation if joining the projections associated with the schema results in the original universal relation.
An intuitive and standard measure of loss entailed by an AJD is the number of redundant tuples generated by the acyclic join. 
In this talk, I will show how the loss associated with an AJD can be characterized 
by an information-theoretic measure, which gives rise to a principled definition of approximation for acyclic schemas. I will then describe how this definition was applied in Maimon: a system for automatically discovering approximate acyclic schemas that hold in the data.  
 
RELATED PAPERS:
1. Mining Approximate Acyclic Schemas from Relations (SIGMOD 2020) (https://dl.acm.org/doi/pdf/10.1145/3318464.3380573)
2. Quantifying the loss of Acyclic Join Dependencies (To Appear in PODS 2023) (https://arxiv.org/pdf/2210.14572.pdf) 
Code for MAIMON can be found here:
https://github.com/orglassman/MAIMON

BIO
Batya Kenig is an assistant professor in the faculty of data and decision sciences at the Technion. Previously, she was a postdoctoral research associate at the database group at the University of Washington. Her research interests include probabilistic reasoning, uncertain data, and enumeration algorithms.

TITLE
Toward a Unified Solution for Reverse Data Management 
 
ABSTRACT
The notion of Reverse Data Management captures various problems in database theory such as view maintenance, deletion propagation, and various interventions for fairness. We study Resilience: a key algorithmic problem that underlies the various forms of reverse data management. Resilience asks the question, "What is the minimal number of tuples to delete from a database in order to remove all answers from a query?" A long-open question is determining those conjunctive queries (CQs) for which this problem can be solved in guaranteed PTIME. We shed new light on this and the related problem of Causal Responsibility by proposing and studying the properties of a unified Integer Linear Programming (ILP) formulation.

Our algorithm is unified in that it can solve both prior studied restrictions (e.g., self-join-free CQs under set semantics that allow a PTIME solution) and new cases (e.g., all CQs under set or bag semantics). It is also unified in that all queries and all instances are treated with the same approach, and the algorithm is guaranteed to terminate in PTIME for the easy cases. Concretely, we prove that for all easy self-join-free CQs, the Linear Programming (LP) relaxation of our encoding is identical to the ILP solution and thus standard ILP solvers are guaranteed to return the solution in PTIME. Our approach opens the door to new variants and new fine-grained analysis: 1) It also works under bag semantics, and we give the first dichotomy result for bags semantics in the problem space. 2) We give a more fine-grained analysis of the complexity of causal responsibility revealing new PTIME cases. 3) We recover easy instances for generally hard queries, such as instances with read-once provenance and instances that become easy because of Functional Dependencies in the data. 4) We solve an open conjecture from PODS 2020 about a structural hardness criterion. 5) Experiments confirm that our results indeed predict the asymptotic running times, and that our universal ILP encoding is at times even faster to solve for the PTIME cases than a prior proposed dedicated flow algorithm. 6) Our technique shifts the perspective from finding PTIME algorithms for open problems to using a unified algorithm and resolving open complexity questions by “merely” proving that the algorithm terminates in PTIME.
 
RELATED PAPERS:
1. Neha Makhija, Wolfgang Gatterbauer. "A Unified Approach for Resilience and Causal Responsibility with Integer Linear Programming (ILP) and LP Relaxations". Arxiv Preprint (https://arxiv.org/abs/2212.08898)

BIO
Neha Makhija is a PhD student at the DataLab at Northeastern University, advised by Prof. Wolfgang Gatterbauer. Her research aims to find efficient algorithms and borders of tractability for problems in reverse data management and minimal representations of knowledge.

TITLE
Data Acquisition for AI on Lakes of Data
 
ABSTRACT
Data for AI is increasingly reliant on the integration of multiple sources – sometimes obtained from public data or data marketplaces. Despite decades of research in data integration and cleaning, we are still not sure how to construct AI-ready datasets – data with relevant features and representative distribution. In this talk, first, I will describe how to discover relevant datasets from large-scale data repositories, by designing (semantic) similarity measures and efficient index structures. Next, I will show how to use ideas from randomized algorithms to tailor a dataset with desired distribution requirements from multiple sources, in order to construct fair datasets. We will also see how to obtain an IID sample over normalized data, to improve the efficiency of model training and approximate query answering. Finally, I will conclude by discussing distribution-aware and human-centric data acquisition.

BIO 
Fatemeh Nargesian is an assistant professor of computer science at the University of Rochester. She obtained her PhD at the University of Toronto. Her research interests are in data management for AI-based data analytics and scientific time-series management. Her work has appeared at top-tier venues including VLDB, SIGMOD, and ICDE, and has won the best demo award of VLDB 2017. 

TITLE:
Commercial-Grade Static Analyzers in Datalog 
 
ABSTRACT:

Software development and quality assurance teams use static analyzers to find bugs and adhere to industry standards. Static analyzers must rapidly adapt to various programming styles and project requirements. In addition, they must scale to process large code bases with adequate precision. Unfortunately, implementing commercial-grade static analyzers is tedious, error-prone, time-consuming, and costly.

This talk will outline the rapid implementation of commercial-grade static analyzers in modern Datalog languages and present some industrial use cases. We will discuss the performance challenges of modern Datalog engines used for implementing static analyzers and techniques to make static analyzers scale and truly declarative.

RELATED PAPERS:

Souffle Papers can be found here but there are also many other Souffle papers out there: https://souffle-lang.github.io/publications

BIO:
 
Bernhard Scholz is a Full Professor in the School of Computer Science at The University of Sydney. His research interests are in Programming Language Research. Prof Scholz is the creator of Souffle, a Datalog language that is mainly used for static program analysis by companies such as Oracle to find security bugs in Java JDK and check the correctness of Amazon’s VPC networks.

Prof Scholz co-founded the SUN Microsystems Labs in Brisbane. He co-initiated the bug-checking tool Parfait, a de facto tool used by thousands of Oracle developers for bug and vulnerability detection in real-world, commercially sized C/C++/Java applications. Prof Scholz introduced the Partitioned Boolean Quadratic Problem (PBQP) for compiler construction with his students. It is employed by compilers such as LLVM for register allocation.

Before joining The University of Sydney, he worked for the Technical University of Vienna and the University of Vienna in academic/research roles. He has held various visiting professorships at the University of Victoria, BC, Canada, Yonsei University, South Korea, at the Sun Microsystems Laboratories, and Oracle Labs. He has taken a leave of absence to work full-time for the blockchain foundation Fantom as a chief research officer. Prof. Scholz’s blockchain research focuses on smart-contract execution, security, and performance.

TITLE
Enhancing Database Applications With Causality
 
ABSTRACT

Causal inference is at the heart of multiple branches of science as means of drawing conclusions and assisting in decision-making. In database research, notions from causal inference have already proved useful in several applications, such as providing explanations for query results. In this talk, I will describe additional useful database applications that can also benefit from a causality-informed approach. In particular, I will present two recent works that incorporate the advantages of causal inference into two well-established database uses: hypothetical reasoning and differentially-private synthetic data generation. I will begin by discussing our work, called HypeR, a framework for hypothetical reasoning in databases that uses a causal inference approach to capture fine-grained hypothetical updates. I will proceed by presenting our recent work, PreFair, a framework for synthetic data generation that is both differentially-private and fair, where fairness is modeled by an adaptation of the causal definition for justifiable fairness.  
 
RELATED PAPERS
 
1. Sainyam Galhotra, Amir Gilad, Sudeepa Roy, Babak Salimi-  HypeR: Hypothetical Reasoning With What-If and How-To Queries Using a Probabilistic Causal Approach- SIGMOD '22(https://dl.acm.org/doi/10.1145/3514221.3526149)
2. David Pujol, Amir Gilad, Ashwin Machanavajjhala - PreFair: Privately Generating Justifiably Fair Synthetic Data- To appear in VLDB '23(https://arxiv.org/abs/2212.10310)

BIO

Amir Gilad is currently a postdoctoral researcher in the Database Group at Duke University and will join The School of Computer Science and Engineering at the Hebrew University as an Assistant Professor at the end of this year. 
Amir has received his Ph.D. in Computer Science from Tel Aviv University. His work focuses on developing tools and algorithms that assist users in understanding and gaining insights into data and the systems that manipulate it. His research relates to classic database tools such as data provenance, as well as natural language processing, causal inference, and privacy. Amir is the recipient of the VLDB best paper award, the SIGMOD research highlight award, and the Google Ph.D. fellowship for Structured Data and Database Management. 

TITLE:
Table Discovery and Integration in Data Lakes: Challenges and Solutions
 
ABSTRACT:
Data scientists use tabular data to support decision-making processes, train machine learning models, perform statistical analysis, and more. The enormous size and heterogeneity of data lakes, together with the fact that metadata may be missing or unreliable, make it challenging for data scientists to find suitable tables for their analyses. In this talk, we will cover recent results that help data scientists efficiently and effectively find tables from the data lake that union with their query table. We will also discuss new results on how to best integrate the discovered tables particularly when the discovered tables may be incomplete and simple union is insufficient.
First, we will present SANTOS, a recent technique that finds tables from data lakes that can union with a query table given by the user. SANTOS introduces a new notion of unionability that considers relationships between columns, together with the semantics of columns, in a principled way. To do so, we present two methods to discover semantic relationships between pairs of columns, using existing and data-driven synthesized knowledge bases. We next present our work on Starmie, which learns the semantics of columns in a natural language domain by using the context of the entire table. Moreover, Starmie is the first to employ the HNSW (Hierarchical Navigable Small World) index to accelerate query processing of table union search.
Lastly, we will cover another recent work called ALITE (Align and Integrate) that integrates the tables that may have been discovered using join, union, or related table search. Specifically, ALITE aligns the columns of discovered tables using holistic schema matching and then applies the Full Disjunction over the matching columns to get an integrated table.
 
RELATED WORK:
1) Aamod Khatiwada, Grace Fan, Roee Shraga, Zixuan Chen, Wolfgang Gatterbauer, Renée J. Miller, Mirek Riedewald: SANTOS: Relationship-based Semantic Table Union Search. To appear in SIGMOD 2023 (arXiv)
2) Grace Fan, Jin Wang, Yuliang Li, Dan Zhang, Renée J. Miller: Semantics-aware Dataset Discovery from Data Lakes with Contextualized Column-based Representation Learning. To appear in PVLDB 2023 (arXiv)
3) Aamod Khatiwada, Roee Shraga, Wolfgang Gatterbauer, Renée J. Miller: Integrating Data Lake Tables. Proc. VLDB Endow. 16(4): 932-945 (2022) (https://www.vldb.org/pvldb/vol16/p932-khatiwada.pdf)
 
BIO:
Grace Fan is a third year PhD student at the Data Lab at Northeastern University, advised by Professor Renée J. Miller. Her research interests include data discovery and data integration in data lakes. 
Aamod Khatiwada is a third year PhD Student at the Data Lab at Northeastern University, advised by Professor Renée J. Miller. His research aims to develop novel techniques and algorithms that can find and integrate open data tables in a principled way.

TITLE
Tensors: An abstraction for data processing
 
ABSTRACT
Deep Learning (DL) has created a growing demand for simpler ways to develop complex models and efficient ways to execute them. Thus, a significant effort has gone into frameworks like PyTorch or TensorFlow to support a variety of DL models and run efficiently and seamlessly over heterogeneous and distributed hardware. Since these frameworks will continue improving given the predominance of DL workloads, it is natural to ask what else can be done with them. This is not a trivial question since these frameworks are based on the efficient implementation of tensors, which are well adapted to DL but, in principle, to nothing else. In this paper, we explore to what extent Tensor Computation Runtimes (TCRs) can support non-ML data processing applications, so that other use cases can take advantage of the investments made on TCRs. In particular, we are interested in graph processing and relational operators, two use cases very different from ML, in high demand, and complement quite well what TCRs can do today. Building on Hummingbird, a recent platform converting traditional machine learning algorithms to tensor computations, we explore how to map selected graph processing and relational operator algorithms into tensor computations. Our vision is supported by the results: our code often outperforms custom-built C++ and CUDA kernels, while massively reducing the development effort, taking advantage of the cross-platform compilation capabilities of TCRs.


 
RELATED PAPERS:
1) He, Dong, et al. "Query processing on tensor computation runtimes." arXiv preprint arXiv:2203.01877 (2022) (https://arxiv.org/abs/2203.01877).
2) Müller, Ingo, et al. "The collection Virtual Machine: an abstraction for multi-frontend multi-backend data analysis." Proceedings of the 16th International Workshop on Data Management on New Hardware. 2020 (https://doi.org/10.1145/3399666.3399911).
3) Koutsoukos, Dimitrios, et al. "Modularis: modular relational analytics over heterogeneous distributed platforms." arXiv preprint arXiv:2004.03488 (2020) (https://arxiv.org/abs/2004.03488).

BIO  
Dimitris Koutsoukos is a PhD student in the Systems Group at ETH Zurich, supervised by Gustavo Alonso and Ana Klimovic. He is broadly interested in database systems and more specifically on how we can execute database workloads on heterogeneous architectures (RDMA, serverless, NVM) and adapt the next generation of DBMSs in the cloud. More recently, he is looking into using compiler infrastructure like MLIR for databases and how we can optimize data layouts for serverless databases. He holds a MSc in Data Science from ETH Zurich. Before coming to ETH, he obtained my undergraduate degree in Electrical and Computer Engineering from the National Technical University of Athens. In the past, he has worked for Meta, Orfium, Maastricht University and he has collaborated with Microsoft Gray Systems Lab.

TITLE
Designing Human-Centered AI Systems for Human-AI Collaboration

ABSTRACT
Human-Centered AI (HCAI) refers to the research effort that aims to design and implement AI techniques to support various human tasks, while taking human needs into consideration and preserving human control. Prior work has focused on human-AI interaction interface design and explainable AI research (XAI). However, despite these fruitful research results, why do many so-called “human-centered” AI systems still fail in the real world? In this talk, I will discuss the human-AI collaboration interaction paradigm, and show how we can learn from human-human collaboration to design and build AI systems that lead to this interaction paradigm. This work serves as a cornerstone towards the ultimate goal of Human-AI Collaboration, where AI and humans can take complementary and indispensable roles to achieve a better outcome and experience.

RELATED PAPERS
* Human-AI Collaboration in Data Science: Exploring Data Scientists’ Perceptions of Automated AI. CSCW 2019. https://arxiv.org/pdf/1909.02309.pdf
* Telling Stories from Computational Notebooks: AI-Assisted Presentation Slides Creation for Presenting Data Science Work. CHI 2022. https://arxiv.org/pdf/2203.11085.pdf
* Model Sketching: Centering Concepts in Early-Stage Machine Learning Model Design. CHI 2023. https://arxiv.org/pdf/2303.02884.pdf

BIO
Dr. Dakuo Wang is an Associate Professor at Northeastern University and Visiting Scholar at Stanford University. His research lies at the intersection of human-computer interaction (HCI), artificial intelligence (AI), and computer-supported team collaboration (CSCW), with a focus on the exploration, development, and evaluation of human-centered AI (HCAI) systems. The overarching research goal is to democratize AI for every person and every organization, so that they can easily access AI and collaborate with AI to accomplish real-world tasks better -- the “human-AI collaboration” paradigm. Before joining Northeastern, Dr.Wang worked as a research lead at IBM Research, and a principal investigator at MIT-IBM Watson AI Lab. He got his Ph.D. from the University of California Irvine (“how people write together now” co-advised by Judith Olson and Gary Olson). He has worked as a designer, researcher, and engineer in the U.S., China, and France. He has served in various organizing committees, program committees, and editorial boards for conferences and journals, and ACM has recognized him as an ACM Distinguished Speaker.

TITLE
Human-AI Techniques to Handle Open-Ended Answers

ABSTRACT
Collecting answers to open-ended questions is essential for many applications.For instance, collecting the names of social influencers is useful for brandmarketing and opinion mining. Another example of an open-ended task is theextraction of rationales from documents for topic classification, which ishelpful in literature review and contextual search. Many efforts have beendedicated to automatically curating such data through ML models. However, thesemethods have limited performance since responding to open-ended questionsrequires intuition, domain knowledge, and reasoning abilities, which are stilllargely missing in state-of-the-art ML methods. In this talk, I will presentseveral human-AI collaborative approaches for handling open-ended data. Overall,our frameworks consist of a human computation component and an AI model thatinteract with each other. The human computation component and the AI model areupdated iteratively, allowing their learning processes to benefit from eachother until an agreement on the quality of the answers is reached. Thus, theinteraction between the human computation component and the AI model isbidirectional, which is fundamental to ensuring the effectiveness of thehuman-AI team.

RELATED PAPERS
- Ines Arous, Jie Yang, Mourad Khayati, andPhilippe Cudre-Mauroux. ”Opencrowd: A human-AI collaborative approach forfinding social influencers via open-ended answers aggregation.” WWW 2020. https://exascale.info/assets/pdf/arous2020www.pdf

- Ines, Arous, JieYang, Mourad Khayati, and Philippe Cudre-Mauroux. ”Peer grading the peerreviews: a dual-role approach for lightening the scholarly paper reviewprocess.” WWW 2021. https://exascale.info/assets/pdf/arous2021www.pdf

- Ines Arous, LjiljanaDolamic, Jie Yang, Akansha Bhardwaj, Giuseppe Cuccu, and PhilippeCudre-Mauroux. ”Marta: Leveraging human rationales for explainable textclassification.” AAAI 2021. https://exascale.info/assets/pdf/arous2021aaai.pdf

BIO
Philippe Cudre-Mauroux is a Full Professor and the Director of the eXascaleInfolab at the University of Fribourg in Switzerland. He received his Ph.D.from the Swiss Federal Institute of Technology EPFL, where he won both theDoctorate Award and the EPFL Press Mention in 2007. Before joining theUniversity of Fribourg, he worked on information management infrastructures atIBM Watson (NY), Microsoft Research Asia and Silicon Valley, and MIT. Herecently won the Verisign Internet Infrastructures Award, a Swiss NationalCenter in Research award, a Google Faculty Research Award, as well as a 2million Euro grant from the European Research Council. His research interestsare in next-generation, Big Data management infrastructures for non-relationaldata and AI. Webpage: http://exascale.info/phil

TITLE
Interpretable and Actionable Data Analysis with Explanations and Causality

ABSTRACT
In current times, data is considered synonymous with knowledge, profit, power, and entertainment, requiring development of new techniques to extract useful information and insights from data. In this talk, I will describe some concepts and techniques in interpretable data analysis from the viewpoint of a database researcher. First, I will talk about our work on explaining query answers, in terms of "intervention" or how changes in the input data changes the output of a query, and "context" or how input data not contributing to the answers of interest can help explain them. In particular, I will describe a recent work on an explanation framework called DPXPlain that preserves Differential Privacy for sensitive data. Then I will talk about true causal inference from observational data without randomized controlled experiments and how database techniques can help with causal inference for large complex data.

RELATED PAPERS

- https://www.vldb.org/pvldb/vol16/p113-tao.pdf

- https://dl.acm.org/doi/abs/10.5555/3546258.3546289

BIO
Sudeepa Roy is an Associate Professor of Computer Science at Duke University. She works broadly in data management, with a focus on foundational aspects of big data analysis, including causality and explanations, data provenance, uncertain data, data repair, query optimization, and database theory. Prior to joining Duke in 2015, she did a postdoc at the University of Washington, and obtained her Ph.D. from the University of Pennsylvania. She is a recipient of an NSF CAREER Award, the Very Large Databases (VLDB) Early Career Research Contributions Award, and a Google Ph.D. fellowship in structured data. She co-directs the Almost Matching Exactly (AME) lab for interpretable causal inference at Duke (https://almost-matching-exactly.github.io/). 

TITLE

Collective Grounding: Applying Database Research to SRLScalability

ABSTRACT

The process of instantiating, or "grounding", a first-order model is a fundamental component of reasoning in logic.
It has been widely studied in the context of theorem proving, database theory, and artificial intelligence.
Within the relational learning community, it is required for instantiating a distribution over possible worlds and is often the key computational bottleneck.
In this work, we investigate how key concepts from relational database theory can be utilized to improve the computational efficiency of the grounding process.
We introduce the notion of collective grounding, which treats logical programs not as a collection of independent rules, but instead as a joint set of interdependent workloads that can be shared.
We introduce the theoretical concept of collective grounding, the components necessary in a collective grounding system, implementations of these components, and how to use database theory to speed up these components.
We demonstrate collective groundings effectiveness on seven popular datasets, and show up to an 80\% reduction in runtime using collective grounding.

RELATED PAPERS

Hinge-Loss Markov Random Fields and Probabilistic Soft Logic
https://linqs.org/publications/#id:bach-jmlr17

A Comparison of Bottom-Up Approaches to Grounding for Templated Markov Random Fields
https://linqs.org/publications/#id:augustine-mlsys18

Tuffy: Scaling up Statistical Inference in Markov Logic Networks using an RDBMS
http://i.stanford.edu/hazy/papers/tuffy-vldb11.pdf

BIO

Eriq Augustine is a doctoral candidate with the LINQS labheaded by Dr. Lise Getoor at the University of California at Santa Cruz.
Eriq works at the intersection of relational information (e.g., logic) and machinelearning with a focus on scalability, data management, and systems engineering.
Before starting at UCSC, Eriq was a lecturer at the California StatePolytechnic University at San Luis Obispo, a senior software developer at GaineSolutions (a master data management startup), a software developer at Google,and a software developer at Netflix.
eriqaugustine.com

TITLE
Semantic Advancement of Data for Data Analytics


ABSTRACT
In this work, we use semantic knowledge sources, such as cross-domain knowledge graphs (KGs) and domain-specific ontologies, to enrich structured data for various anaytics applications. By enriching our understanding of the underlying data with semantics brought in from external ontologies and KGs, we can better interpret the data as well as the queries to answer more questions, provide more complete answers, and deal with entity disambiguation. To semantically enrich the data with external knowledge sources, we need to find the correspondences between the structured data and the entities in the cross-domain KGs and/or the domain-specific ontologies. In this presentation, we break this problem into several steps, and provide solutions for each step. We showcase the practical value of semantic enrichment of data using our proposed techniques in entity disambiguation, natural language querying and conversational interfaces to data, query relaxation, with promising initial results.


RELATED PAPERS

1. J. Hao, C. Lei, V. Efthymiou, A. Quamar, F. Özcan, Y. Sun, and W. Wang, “MEDTO: Medical Data to Ontology Matching Using Hybrid Graph Neural Networks”, KDD 2021
2. A. Vretinaris, C. Lei, V. Efthymiou, X. Qin, and F. Özcan, “Medical Entity Disambiguation Using Graph Neural Networks”, SIGMOD 2021
3. C. Lei, V. Efthymiou, R.Geis, and F. Özcan, “Expanding Query Answers on Medical Knowledge Bases”, EDBT 2020 (industrial)
4. A. Quamar, F. Özcan, D. Miller, R. Moore, R. Niegus, and J. Kreulen, “Conversational BI: An Ontology-Driven Conversation System for Business Intelligence Applications”, PVLDB Volume 13, No 12, August 2020
5. J. Sen,  C. Lei,  A. Quamar, F. Özcan, V. Efthymiou, A. Dalmia, G. Stager,  A. Mittal,  D. Saha, K. Sankaranarayanan, "ATHENA++: Natural Language Querying for Complex Nested SQL Queries", PVLDB Volume 13, No 11, August 2020

BIO
Fatma Özcan is a Principal Engineer at Systems Research@Google. Before that, she was a Distinguished Research Staff Member and a senior manager at IBM Almaden Research Center. Her current research focuses on platforms and infra-structure for large-scale data analysis, query processing and optimization of semi-structured data, and democratizing analytics via NLQ and conversational interfaces to data. Dr Özcan got her PhD degree in computer science from University of Maryland, College Park, and her BSc degree in computer engineering from METU, Ankara. She has over 21 years of experience in industrial research, and has delivered core technologies into IBM products. She has been a contributor to various SQL standards, including SQL/XML, SQL/JSON and SQL/PTF. She is the co-author of the book "Heterogeneous Agent Systems", and co-author of several conference papers and patents. She received the VLDB Women in Database Research Award in 2022. She is an ACM Distinguished Member, and the vice chair of ACM SIGMOD. She has been serving on the board of directors of CRA (ComputingResearch Association) since 2020, and is a steering committee member of the CRA-Industry.

TITLE
Taming Large Intermediate Results for Joins over Graph-Structured Relations: A System Perspective

ABSTRACT
Querying graph structured relations, i.e., those with many-to-many relationships between entities, is ubiquitous and integral to a wide range of analytical applications such as recommendations on social networks and fraud detection in financial transactional networks. These queries are heavy on joins and lead to an explosion in the size of intermediate results. In this talk, I cover the details of the integration of two important modern techniques into an analytical DBMS to optimize for many-to-many joins. These two techniques are: (i) novel worst-case-optimal join algorithms that adopt column-at-a-time over the traditional table-at-a-time join paradigm; and (ii) factorized query processing, which is a technique to compress the intermediate and final results of queries. I will present a comprehensive query processor design and implementation that goes over various research questions tackled to make the integration possible and these techniques practical. Our query processor is capable of having a robust query plan space with many efficient plans that lead to orders of magnitude speedups. 


RELATED PAPERS
1) Optimizing One-time and Continuous Subgraph Queries using Worst-Case Optimal Joins [PDF] [Code] [BibTeX]
Amine Mhedhbi, Chathura Kankanamge, Semih Salihoglu
In ACM Transactions on Database Systems (TODS) 2021.
 
2) Columnar Storage and List-based Processing for Graph Database Management Systems [PDF] [Code] [BibTeX]
Pranjal Gupta, Amine Mhedhbi, Semih Salihoglu
In Proceedings of the VLDB Endowment (PVLDB) 2021.

BIO
Amine Mhedhbi is a Ph.D. student at the University of Waterloo. His research focuses on building data systems capable of efficient processing of graph data. To that end, he led the design and implementation of the GraphflowDB system to rethink core database components. He received the VLDB Best Paper Award in 2018 and his research is supported by a Microsoft Ph.D. Research Fellowship.

TITLE
Fairness in Ranking: from values to technical choices and back

ABSTRACT
Algorithmic rankers take a collection of candidates as input and produce a ranking (permutation) of the candidates as output. In the past few years, there has been much work on incorporating fairness requirements into rankers, with contributions from the data management, algorithms, information retrieval, and recommender systems communities. Many fair ranking methods establish their own understanding of fairness, yet they stop short of categorizing themselves with respect to the normative frameworks they embed. This makes it difficult to understand what values these methods encode, in what specific scenarios they are applicable, and how they relate to one another. Further, methods differ on which technical choices they make, including, notably, their data representation choices. These technical differences are not value-neutral, and have profound implications on the applicability of the methods. In my talk, I will give an overview of the field of fairness in raking, offering a perspective that connects formalizations and algorithmic approaches across subfields. My perspective will be based on the interplay between the value frameworks that motivate specific fairness-enhancing interventions, and the technical choices that impact the properties of the methods and of their results. I will compare and contrast several representative fair ranking methods, and will provide concrete recommendations for those wishing to incorporate fairness objectives into algorithmic rankers. My talk will be based on a two-part survey that I co-authored with Meike Zehlike and Ke Yang.

RELATED PAPERS
The survey was recently published in ACM Computing Surveys, and is available at these links:
1. Part I https://dl.acm.org/doi/abs/10.1145/3533379
2. Part II https://dl.acm.org/doi/10.1145/3533380

BIO
Julia Stoyanovich is an Institute Associate Professor of Computer Science & Engineering at the Tandon School of Engineering, Associate Professor of Data Science at the Center for Data Science, and Director of the Center for Responsible AI at New York University (NYU). Her research focuses on responsible data management and analysis: on operationalizing fairness, diversity, transparency, and data protection in all stages of the data science lifecycle. She established the "Data, Responsibly" consortium and served on the New York City Automated Decision Systems Task Force, by mayoral appointment. Julia developed and has been teaching courses on Responsible Data Science at NYU, and is a co-creator of an award-winning comic book series on this topic. In addition to data ethics, Julia works on the management and analysis of preference and voting data, and on querying large evolving graphs. She holds M.S. and Ph.D. degrees in Computer Science from Columbia University, and a B.S. in Computer Science and in Mathematics & Statistics from the University of Massachusetts at Amherst. She is a recipient of an NSF CAREER award and a Senior Member of the ACM.

TITLE
Interaction History for Building Human Data Interfaces

ABSTRACT
History provides context for the present. In the same way, past user interactions provide context for present explorations. This thesis investigates ways to reify user interaction history to address emerging challenges in the design and programming of human data interfaces.
We leverage interaction history in three different but connected ways. First is to enhance the design of interactions with delays, such as when working with remote databases. We use interaction history as a visual anchor to facilitate concurrent interactions, which ameliorate the cognitive burdens caused by delays. Second is to facilitate the programming of interactive visualizations involving asynchronous communication with remote databases. We capture event histories as a first-class programming construct, allowing the developer to declaratively specify what data to compute and how to update the state of the user interface. This way, developers avoid the low-level details of accessing remote data and coordinating events. The third and last way is to use interaction history to create a bridge between interaction design and programming. We capture and reify interaction history in both computational notebooks and interactive visualizations. Affordances on these reified histories help data scientists move fluidly between the two mediums.

RELATED PAPERS
https://ieeexplore.ieee.org/abstract/document/9331306
https://ieeexplore.ieee.org/abstract/document/9552893
https://dl.acm.org/doi/pdf/10.1145/3379337.3415851

BIO
Yifan Wu graduated from UC Berkeley in 2020, advised by Joe Hellerstein, and in collaboration with Arvind Satyanarayan, Eugene Wu, and Remco Chang. After graduation, Yifan founded and left a MLOps startup.

TITLE
Leveraging tensor train decomposition to deal with uncertainty in  multidimensional data

ABSTRACT
In many domains, multidimensional data are encoded as multi-dimensional arrays, or Tensors, and traditional data mining techniques might be limited due to the curse of dimensionality – the memorization space is indeed exponential in the number of dimensions. Tensor decomposition techniques alleviate this issue, in that they allow to represent tensors in terms of a reduced number of parameters, thus reducing the space required to describe the relevant information. Commonly used tensor decomposition algorithms include CP-decomposition (which seeks a diagonal core) and Tucker-decomposition (which seeks a dense core). Naturally, Tucker maintains more information, but due to the denseness of the core, it also is subject to exponential memory growth with the number of tensor modes. Tensor train (TT) decomposition addresses this problem by seeking a sequence of three-mode cores.
I will present the main ideas underlying tensor train decomposition and show how its properties can be leveraged to deal with uncertain information (both uncertainty modelled probabilistically, and uncertainty in terms of intervals, ranges, of possible values). Moreover, I will show how knowledge about data characteristics and relationships potentially defined on the tensor modes can be leveraged to guide the decomposition order, to improve the effectiveness of the decomposition.

RELATED PAPERS
Mao-Lin Li, K. Selçuk Candan, Maria Luisa Sapino
GTT: Leveraging data characteristics for guiding the tensor train decomposition.
Inf. Syst. 108: 102047 (2022)
https://scholar.google.com/scholar?cluster=9452783075522390103
 
Francesco Di Mauro; K. Selcuk Candan; Maria Luisa Sapino
Tensor-Train Decomposition in Presence of Interval-Valued Data
IEEE Transactions on Knowledge and Data Engineering, 2021
https://scholar.google.com/scholar?cluster=9373470206136825
 
Xinsheng Li, K. Selçuk Candan, Maria Luisa Sapino:
Noise Adaptive Tensor Train Decomposition for Low-Rank Embedding of Noisy Data.
SISAP 2020: 203-217
https://scholar.google.com/scholar?cluster=8066453498002786335

BIO

Maria Luisa Sapino is a Professor at the University of Torino and, since 2006, Adjunct Professor at the School of Computing and Augmented Intelligence, SCAI, at the Arizona State University. At ASU she is also affiliated with the Center for Assured and Scalable Data Engineering (CASCADE). Her initial contributions to computer science were in the area of logic programming and artificial intelligence, specifically in the semantics of negation in logic programming, and in the abductive extensions of logic programs. Since mid-90s she has been applying these techniques to the challenges associated with data security and access control, and with heterogeneous data management. Her research contributions are in the area of  data management and analysis, with strong emphasis on tackling the so called "Big Data challenges", including aspects related to the development of efficient techniques for tensor based big data analysis,  pattern detection in voluminous and heterogeneous data collections,  with special focus on various aspects related to indexing, classification and querying of (possibly multivariate) time series. Maria Luisa Sapino co-authored more than 100 research papers, most of them published on top ranked journals and conference proceedings. Her recent research includes interdisciplinary projects focusing on securing and optimizing health and infrastructure (study of the infectious disease propagation, study of real-time evacuation solutions in case of disaster management) as well as cyber-physical systems (building energy systems) that can benefit from "smart data oriented" fundamental technological innovations.

TITLE:
HumanAL: Calibrating Human Matching Beyond a Single Task

ABSTRACT
This work offers a novel view on the use of human input as labels, acknowledging that humans may err. We build a behavioral profile for human annotators which is used as a feature representation of the provided input. We show that by utilizing black-box machine learning, we can take into account human behavior and calibrate their input to improve the labeling quality. To support our claims and provide a proof-of-concept, we experiment with three different matching tasks, namely, schema matching, entity matching and text matching. Our empirical evaluation suggests that the method can improve the quality of gathered labels in multiple settings including cross-domain (across different matching tasks).

RELEVANT PAPER:
https://arxiv.org/abs/2205.03209

SHORT BIO
Roee Shraga is a Postdoctoral fellow at the Khoury College of Computer Science at Northeastern University in Boston. He received his PhD degree in 2020 in the area of Data Science. Roee has published more than a dozen papers in leading journals and conferences and is a recipient of several PhD fellowships including the Miriam and Aaron Gutwirth Memorial Fellowship (2020).
https://sites.google.com/view/roee-shraga/

TITLE
Data Systems for Economic Discovery

ABSTRACT
Macroeconomics resembles a data science task that predates modern data science, and arguably predates digital computing as well. It is possible to make economics easier and to make novel discovery through application of modern data systems and data science methods. This talk will describe several areas for data management to make contributions to economics, in particular recent work on the crucial task of creating high-quality price indices, which describe inflation and consumption.

RELEVANT PAPERS
https://www.nber.org/system/files/working_papers/w20010/w20010.pdf
http://www-personal.umich.edu/~shapiro/papers/Price_Quantity_Scale.pdf
https://web.eecs.umich.edu/~michjc/papers/p145-antenucci.pdf

BIO
Michael Cafarella is a Principal Research Scientist at MIT CSAIL. His research interests include databases, information extraction, data integration, and data mining. He has published extensively in venues such as SIGMOD, VLDB, and elsewhere. Mike received his PhD from the University of Washington, Seattle, in 2009 with advisors Oren Etzioni and Dan Suciu. From 2010 to 2020, he was on the faculty of Computer Science and Engineering at the University of Michigan. His academic awards include the NSF CAREER award, the Sloan Research Fellowship, and the 2018 VLDB Ten-Year Best Paper award. In addition to his academic work, he costarted (with Doug Cutting) the Hadoop open-source project, which is widely used at Facebook, Yahoo!, and elsewhere. In 2015 he cofounded (with Chris Re and Feng Niu) Lattice Data, Inc, which is now part of Apple.
https://www.csail.mit.edu/person/michael-cafarella

TITLE
Scalable Dynamic Graph Summarization

ABSTRACT
Large-scale dynamic interaction graphs can be challenging to process and store, due to their size and the continuous change of communication patterns between nodes. In this work, we address the problem of summarizing large-scale dynamic graphs, while maintaining the evolution of their structure and interactions. Our approach is based on grouping the nodes of the graph in supernodes according to their connectivity and communication patterns. The resulting summary graph preserves the information about the evolution of the graph within a time window. We propose two online algorithms for summarizing this type of graphs. Our baseline algorithm kC based on clustering is fast but rather memory expensive. The second method we propose, named /LC, reduces the memory requirements by introducing an intermediate step that keeps statistics of the clustering of the previous rounds. Our algorithms are distributed by design, and we implement them over the Apache Spark framework, so as to address the problem of scalability for large-scale graphs and massive streams. We apply our methods to several dynamic graphs, and show that we can efficiently use the summary graphs to answer temporal and probabilistic graph queries.

RELATED PAPERS
1. https://link.springer.com/article/10.1007/s10618-016-0468-8
2. https://www.sciencedirect.com/science/article/pii/B9780127224428500161
3. https://cs-people.bu.edu/evimaria/papers/Social-net.pdf

BIO
Ioanna Tsalouchidou received the PhD degree from the Department of Information and Communication Technologies, University Pompeu Fabra, in 2018. She is currently working at Redshift AWS as a software engineer. Her main interests are data bases, geospatial systems, spatial algorithms and data mining, with an emphasis on large-scale and dynamic graphs.
https://www.amazon.science/author/ioanna-tsalouchidou

TITLE
Fantastic Knowledge Graph Embeddings and How to Find the Right Space for Them:
The Quest for Hybrid Logic- and ML-based Reasoning on Knowledge Graphs

ABSTRACT
In this talk, we are going to start by exploring “fantastic knowledge graph embeddings and how to find the right space for them” [1]. The starting point of our quest, namely that of hybrid logic- and ML-based reasoning in Knowledge Graphs (KGs), is simple:
During the last few years, a large number of KG embedding (KGE) models have been devised for handling machine learning problems for KGs. Interestingly, many models capable of inferring relational patterns, such as symmetry or transitivity, show lower performance in practice than those not able to, with the factors contributing to this largely unknown. In this part, we are going to show how to solve this puzzle, and practically improve models [1].
This brings us to the second part of our quest: considering e.g. symmetry or transitivity mentioned above, we of course know reasoning methodologies that are very effective at handling them, namely logic-based KG reasoners (building upon the scalability of techniques from the data management community). Yet can we bring KGE-based and logic-based methods together in meaningful ways? This is a challenge that a large part of our community is currently looking at, and we present interesting parts of this quest including how to improve logic-based reasoning using embeddings in the Vadalog system [2].
We round up the talk by looking at the general picture of Knowledge Graph Management Systems (KGMSs), with a focus on the Vadalog system, and its various applications including anti-money laundering, hostile takeover prevention during crises, etc., as time allows [3].

RELATED PAPERS
1. ”Fantastic Knowledge Graph Embeddings and How to Find the Right Space for Them”
https://link.springer.com/chapter/10.1007/978-3-030-62419-4_25
2. http://ceur-ws.org/Vol-3135/EcoFinKG_2022_paper8.pdf
3. https://www.sciencedirect.com/science/article/abs/pii/S0306437920300351

BIO
Emanuel Sallinger is Assistant Professor at TU Wien and Departmental Lecturer at Oxford University. He leads the Knowledge Graph Lab at TU Wien, the SIG Knowledge Graphs of the Center for AI and ML, and is a member of the Databases and AI group (DBAI). At Oxford University, he is lecturer of the courses “Knowledge Graphs” and “Database Design”. Before joining TU Wien, he directed the VADA (Value-Added Data Systems and Architecture) Laboratory at Oxford University for over six years. The Knowledge Graph Lab (formally “Scalable Reasoning in Knowledge Graphs”) is a WWTF Vienna Research Group, an ERC-sized funding program for creating long-term research groups. The VADA project was an EPSRC programme grant under Georg Gottlob, bringing together the universities of Oxford (directed by Emanuel Sallinger), Manchester (headed by Norman Paton), and Edinburgh (headed by Leonid Libkin).
https://kg.dbai.tuwien.ac.at/

TITLE
RAISON DATA: Some Perspectives on the Intersection of Data, AI, and Logic

ABSTRACT
In the RAISON DATA project, we investigate possible interactions between data, AI, and logic. After a brief overview of motivations and goals for the project, followed by the presentation of two recent publications and ongoing follow-up work.

Part I: Deriving value and insights from data is a central task in industry and extensively studied in data management research. While most work in this area has focused on analysis using complex Machine Learning models and other forms of statistical analysis, there is a renewed interest in methods for deriving symbolic rather than statistical knowledge from data. We investigate the complexity of inductively (exactly) learning guarded rules from examples and discuss ongoing work on applying outcomes from this analysis to relational databases.

Part II: Current trends in data management introduce various data from sources that label data points with a degree of certainty. Common examples include data obtained by crowd-sourced data, AI-generated observations, and KG-completion/link detection techniques. These developments raise new challenges for data management and the need for formalisms that can account for degrees of certainty in data. To this end, we investigate Datalog with fuzzy logic semantics. In the accompanying paper, we show the possibility of semantics based on linear programming for one variant of fuzzy logic Datalog and how this provides one way of introducing fuzzy Datalog+/-. Additionally, we discuss ongoing work on extending the framework from the paper and connections to other recent developments in rule-based reasoning.

RELATED PAPERS
* On the Complexity of Inductively Learning Guarded Clauses (AAAI'22)
https://www.aaai.org/AAAI22Papers/AAAI-1260.DraghiciA.pdf
* MV-Datalog+-: Effective Rule-based Reasoning with Uncertain Observations (ICLP'22)
https://arxiv.org/abs/2202.01718

BIO
Matthias Lanzinger is a senior postdoctoral researcher at the University of Oxford. He obtained his Ph.D. from TU Wien under the supervision of Reinhard Pichler. His main research interests are the complexity of reasoning, with a focus on database query languages, as well as interactions between logic and machine learning systems.
https://www.jesus.ox.ac.uk/about-jesus-college/our-community/people/dr-matthias-lanzinger/

TITLE
Explainable Checking for Factual Claims

ABSTRACT
Misinformation is an important problem but fact checkers are overwhelmed by the amount of false content that is produced online every day. To support fact checkers in their efforts, we are creating data driven verification methods that assess textual claims and explain their decisions. However, using structured data for facts expressed in natural language is a complex task. In this talk, we first discuss how logical rules can be used to verify claims in a checking task modeled as a probabilistic inference problem. We show that this form of reasoning can also be “learned” by transformer-based language models. Experiments show that both methods enable the labeling of claims with interpretable explanations.

RELATED WORK
* Naser Ahmadi, Joohyung Lee, Paolo Papotti, Mohammed Saeed:
Explainable Fact Checking with Probabilistic Answer Set Programming. TTO 2019
https://arxiv.org/pdf/1906.09198
* Mohammed Saeed, Naser Ahmadi, Preslav Nakov, Paolo Papotti:
RuleBERT: Teaching Soft Rules to Pre-Trained Language Models. EMNLP (1) 2021: 1460-1476
https://arxiv.org/pdf/2109.13006

BIO
Paolo Papotti is an Associate Professor at EURECOM, France since 2017. He got his PhD from Roma Tre University (Italy) in 2007, and had research positions at the Qatar Computing Research Institute (Qatar) and Arizona State University (USA). His research is focused on data integration and information quality. He has authored more than 100 publications and his work has been recognized with two "Best of the Conference" citations (SIGMOD 2009, VLDB 2016), two best demo paper awards at SIGMOD (2015, 2022), and two Google Faculty Research Awards (2016, 2020). He is associate editor for the ACM Journal of Data and Information Quality (JDIQ).
https://www.eurecom.fr/~papotti/

TITLE
Benchmarking graph learning algorithms with synthetic data

ABSTRACT
Graph learning algorithms have attained state-of-the-art performance on many graph analysis tasks such as node classification, link prediction, and clustering. It has, however, become hard to track the field's burgeoning progress. One reason is due to the very small number of datasets used in practice to benchmark the performance of graph learning algorithms. This shockingly small sample size (~10) allows for only limited scientific insight into the problem. In this talk, we try to address this deficiency with synthetic data. We generate synthetic graphs and study the behaviour of graph learning algorithms in controlled scenarios. We develop a fully-featured synthetic graph generator that allows deep inspection of different models. We argue that synthetic graph generation allows for thorough investigation of algorithms and provides more insights than overfitting on three citation datasets. In the case study, we show how our framework provides insight into unsupervised and supervised graph neural network models. In this talk, we will explore the benefits, challenges, and insights from synthetic data benchmarking.

RELATED PAPERS
* Tsitsulin, Rozemberczki, Palowitch, Perozzi: Synthetic Graph Generation to Benchmark Graph Learning. 2022
https://arxiv.org/abs/2204.01376
* Palowitch, Tsitsulin, Mayer, Perozzi: GraphWorld -- fake graphs bring real insights for GNNs. 2022.
https://arxiv.org/abs/2203.00112
* Tsitsulin, Palowitch, Perozzi, Mueller: Graph Clustering with Graph Neural Networks. 2020
https://arxiv.org/abs/2006.16904

BIO
Anton Tsitsulin has been a research scientist at Google's graph mining team since 2021. His research is in scalable graph learning algorithms, particularly in unsupervised learning and axiomatic representations. Before joining Google, he received his PhD degree from the university of Bonn advised by Prof. Emmanuel Müller.
http://tsitsul.in/

TITLE
Datalog with Negation and Monotonicity

ABSTRACT
Positive Datalog has several nice properties that are lost when the language is extended with negation. One example is that fixpoints of positive Datalog programs are robust w.r.t. the order in which facts are inserted, which facilitates efficient evaluation of such programs in distributed environments. A natural question to ask, given a (stratified) Datalog program with negation, is whether an equivalent positive Datalog program exists.

In this context, it is known that positive Datalog can express only a strict subset of the monotone queries, yet the exact relationship between the positive and monotone fragments of semi-positive and stratified Datalog was previously left open. In this paper, we complete the picture by showing that monotone queries expressible in semi-positive Datalog exist which are not expressible in positive Datalog. To provide additional insight into this gap, we also characterize a large class of semi-positive Datalog programs for which the dichotomy ‘monotone if and only if rewritable to positive Datalog’ holds. Finally, we give best-effort techniques to reduce the amount of negation that is exhibited by a program, even if the program is not monotone.

RELATED WORK
* Ketsman, Koch. Datalog with Negation. ICDT 2020.
https://drops.dagstuhl.de/opus/volltexte/2020/11943/pdf/LIPIcs-ICDT-2020-19.pdf
* Ameloot, Ketsman, Neven, Zinn. Weaker Forms of Monotonicity for Declarative Networking: A More Fine-Grained Answer to the CALM-Conjecture, TODS 2016.
https://doi.org/10.1145/2809784

BIO
Bas Ketsman is assistant professor at the Vrije Universiteit Brussel (VUB). His research focuses on formal aspects of large-scale data systems. He obtained his Ph.D. from Hasselt University in 2017 under the advice of Frank Neven and was a postdoctoral researcher at the Swiss Federal Institute of Technology in Lausanne (EPFL) where he worked with Christoph Koch. Bas' research has been awarded best paper awards at the ACM PODS Conference (the year 2014 and 2015), the 2018 European Association for Theoretical Computer Science distinguished dissertation award, and also the 2019 SIGMOD Jim Gray Doctoral Dissertation Award Honorable Mention.
https://www.basketsman.com/

TITLE
Scalable Signal Reconstruction for a Broad Range of Applications

ABSTRACT
Signal reconstruction problem (SRP) is an important opti- mization problem where the objective is to identify a solu- tion to an underdetermined system of linear equations that is closest to a given prior. It has a substantial number of applications in diverse areas, such as network traffic engineering, medical image reconstruction, acoustics, astronomy, and many more. Unfortunately, most of the common approaches for solving SRP do not scale to large problem sizes. We propose a novel and scalable algorithm for solving this critical problem. Specifically, we make four major contributions. First, we propose a dual formulation of the problem and develop the Direct algorithm that is significantly more efficient than the state of the art. Second, we show how adapting database techniques developed for scalable similarity joins provides a substantial speedup over Direct. Third, we describe several practical techniques that allow our algorithm to scale—on a single machine—to settings that are orders of magnitude larger than previously studied. Finally, we use the database techniques of materialization and reuse to extend our result to dynamic settings where the input to the SRP changes. Extensive experiments on real- world and synthetic data confirm the efficiency, effective- ness, and scalability of our proposal.

RELATED PAPERS
* Asudeh, Abolfazl, et al. Scalable algorithms for signal reconstruction by leveraging similarity joins. The VLDB Journal 29.2 (2020): 681-707.
https://link.springer.com/article/10.1007/s00778-019-00562-z
* Abolfazl Asudeh, et al. Scalable Signal Reconstruction for a Broad Range of Applications. Communications of the ACM (CACM), Vol. 64(2), pages 106–115, 2021, ACM.
https://dl.acm.org/doi/pdf/10.1145/3441689
* Ives, Zachary G. "Technical perspective: Solving the signal reconstruction problem at scale." Communications of the ACM 64.2 (2021): 105-105.
https://dl.acm.org/doi/pdf/10.1145/3441688

BIO
Abolfazl Asudeh is an assistant professor of Computer Science at the University of Illinois Chicago and the director of Innovative Data Exploration Laboratory (InDeX Lab). He is a VLDB Ambassador, and a regular PC member of Database flagship venues.
His research spans to different aspects of Big Data and Data Science, including Data Management, Information Retrieval, and Data Mining, for which he designs efficient, accurate, and scalable algorithmic solutions.
Data Equity Systems, Algorithmic Fairness, and Data-centric AI are his major focus in research. His research interest also includes: Ranking, Social Networks Analysis, Large-Scale Computation on Limited Resources, Computational Fact Checking, Data management for Machine Learning, Web Databases, and applied Computational Geometry.
https://asudeh.github.io/

TITLE
Constructing Certified Neural Networks for Computer Systems

ABSTRACT
Deep learning and neural networks are powerful tools. Applying them in computer systems---operating systems, databases, and networked systems---attracts much attention. However, neural networks are complicated black boxes that sometimes produce unexpected results. It is risky to use networks with uncertain behaviors in computer systems that require strict safety rules.

To tame networks' uncertainty, we introduce Ouroboros, a system that constructs certified neural networks. Certified neural networks are neural networks that satisfy user-defined safety properties, named specifications. Ouroboros achieves this through a training-verification loop that combines deep learning training and neural network verification. In addition, Ouroboros supports new categories of specifications that enable developers to specify safety properties for systems applications. With Ouroboros, system developers can have faith in their neural networks.

RELATED WORK
* APSys'21: http://naizhengtan.github.io/doc/papers/building21tan.pdf

SHORT BIO
Cheng Tan is an assistant professor of Khoury College of Computer Sciences at Northeastern University. His research interests are in systems and security, focusing on building verifiable outsourced services and certified neural networks for systems. His work has won the SOSP’17 best paper award and Janet Fabri Prize for Outstanding Dissertation.
https://naizhengtan.github.io/

TITLE
Algorithms for Solving The Constrained Most Probable Explanation Problem

ABSTRACT
Recently there has been growing interest in developing probabilistic models that are robust to minor perturbations in input and are explainable in that they are able to explain why they made a particular decision to a user. In this talk, I will describe a new unifying optimization task called constrained most probable explanation (CMPE), and show that the two aforementioned tasks, making models robust and explainable, can be reduced to CMPE. I will show that CMPE is strongly NP-hard in general on arbitrary probabilistic models, but only weakly NP-hard on probabilistic models having small k-separators (a sub-class of tractable models that admit poly-time marginal inference). The main virtue of this weakly NP-hard property is that we can leverage it to derive efficient cutset sampling and local search approximations for lower bounding the optimal value of CMPE. For upper bounding, I will present efficient approaches that combine graph-based partitioning techniques with approximations developed in the literature on knapsack problems. These upper bounding techniques are guaranteed to be better than linear programming based bounds on probabilistic models that admit tractable marginal inference and have much smaller computational complexity. I will end my talk by presenting experimental results and applications as well as avenues for future work.
(Joint work with Sara Rouhani, Rohith Peddi and Tahrima Rahman)

RELEATED PAPERS
* Tahrima Rahman, Sara Rouhani, and Vibhav Gogate. Novel upper bounds for the constrained most probable explanation task. NeurIPS, 2021
https://proceedings.neurips.cc/paper/2021/file/4fc7e9c4df30aafd8b7e1ab324f27712-Paper.pdf
* Sara Rouhani, Tahrima Rahman, and Vibhav Gogate. A novel approach for constrained optimization in graphical models. NeurIPS, 2020
https://proceedings.neurips.cc/paper/2020/file/8ab9bb97ce35080338be74dc6375e0ed-Paper.pdf
* Sara Rouhani, Tahrima Rahman, and Vibhav Gogate. Algorithms for the nearest assignment problem. IJCAI, pages 5096-5102, 2018
https://www.ijcai.org/proceedings/2018/0707.pdf

BIO
Vibhav Gogate is an Associate Professor in the Computer Science Department at the University of Texas at Dallas. He got his Ph.D. at University of California, Irvine in 2009 and then did a two-year post-doc at University of Washington. His research interests are in AI, machine learning and data mining. His ongoing focus is on probabilistic graphical models; their first-order logic based extensions such as Markov logic and probabilistic programming; tractable probabilistic models and explainable AI. He is a recipient of the national science foundation CAREER award and the co-winner of 2010 and 2012 UAI inference competitions.
https://cs.utdallas.edu/people/faculty/gogate-vibhav/

TITLE
Adding Data Management to Orleans – A Journey

ABSTRACT
In this talk, I'll describe my work over eight years to add database features to the Orleans object-oriented programming framework: replication, geo-distribution, transactions, and indexing. The challenge is how to do it when storage is a plug-in service that you don’t control. In this talk, I’ll describe the journey, summarizing the main technical ideas and recounting the ups and downs of a friendly collaboration with a product group. Along the way, I’ll discuss where database system research problems come from and the special challenges of industrial research.

BIO
Philip A. Bernstein is a Distinguished Scientist at Microsoft Research. Over the past 40 years, he has been a product architect at Microsoft and Digital Equipment Corp., a professor at Harvard University, and a VP Software at Sequoia Systems. From 1994-1996 he was architect of Microsoft Repository, a metadata store for Visual Studio and SQL Server. He has published over 150 papers and two books on the theory and implementation of database systems, especially on transaction processing and data integration, as well as many other database topics. He is an ACM Fellow, a winner of the ACM SIGMOD Innovations Award, a member of the Washington State Academy of Sciences, and a member of the National Academy of Engineering. He received a B.S. degree from Cornell and M.Sc. and Ph.D. from University of Toronto.

TITLE
A.M.B.R.O.S.I.A. - Conferring Immortality on Distributed Applications

ABSTRACT
When writing today's distributed programs, which frequently span both devices and cloud services, programmers are faced with complex decisions and coding tasks around coping with failure, especially when these distributed components are stateful. If their application can be cast as pure data processing, they benefit from the past 40-50 years of work from the database community, which has shown how declarative database systems can completely isolate the developer from the possibility of failure in a performant manner. Unfortunately, while there have been some attempts at bringing similar functionality into the more general distributed programming space, a compelling general-purpose system must handle non-determinism, be performant, support a variety of machine types with varying resiliency goals, and be language agnostic, allowing distributed components written in different languages to communicate. This talk describes the first system, publicly available on GitHub, called Ambrosia, to satisfy all these requirements. We coin the term "virtual resiliency", analogous to virtual memory, for the platform feature which allows failure oblivious code to run in a failure resilient manner. We also introduce a programming construct, the "impulse", which resiliently handles non-deterministic information originating from outside the resilient component. Of further interest to our community is the effective reapplication of much database performance optimization technology to make Ambrosia more performant than many of today's non-resilient cloud solutions.

RELATED LINKS
* https://github.com/Microsoft/AMBROSIA

* http://www.vldb.org/pvldb/vol13/p588-goldstein.pdf/

BIO
Over the last 20 years, I have worked at Microsoft in a combination of research and product roles. In particular, I've spent about 20 years as a researcher at MSR, doing fundamental research in streaming, big data processing, databases, and distributed computing. My style of working is to attack difficult problems, and through fundamental understanding and insight, create new artifacts that enable important problems to be solved in better ways. For instance, my work on streaming data processing enabled people with real time data processing problems to specify their processing logic in new, powerful ways, and also resulted in an artifact called Trill, which was orders of magnitude more performant than anything which preceded it. Within the academic community, I have published many papers, some with best paper awards (e.g. Best Paper Award at ICDE 2012), and two with test of time awards (e.g. SIGMOD 2011 Test of Time award and ICDT 2018 Test of Time award), and have also taken many organizational roles in database conferences. My research has also had significant impact on many Microsoft products, including SQL Server, Office, Windows, Bing, and Halo, as well as leading to the creation of entirely new products like Microsoft StreamInsight, Azure Streaming Analytics, Trill, and most recently, Ambrosia. I spent 5 years building Microsoft StreamInsight, serving as a founder and architect for the product. Trill has become the de-facto standard for temporal and stream data processing within Microsoft, and years after creation, is still the most expressive and performant general purpose stream data processor in the world. I am also an inventor of 30+ patents.

TITLE
Large-Scale Graph Algorithms: Massively Parallel Computation

ABSTRACT
Graphs today are massive. Social network graphs, the web graph, and models of the brain are just a few examples of graphs with billions of vertices and trillions of edges. Graphs of this sheer size do not fit a single machine's memory, invalidating many assumptions of traditional algorithms. Massively Parallel Computation (à la MapReduce) is a popular method of processing such large-scale problems which has been extremely successful in practice. The idea is to distribute the workload into several machines running in parallel that communicate over rounds. In this talk, I will give an introduction to two algorithms for maximal matching and graph connectivity in the MPC model. The maximal matching algorithm takes O(log log n) rounds, which exponentially improves upon previous logarithmic-round algorithms known from the 1980s. I will also present a conditionally near-optimal algorithm for graph connectivity in this model.

BIO
Soheil Behnezhad is currently a Motwani Postdoctoral Fellow at Stanford and will join the Khoury College of Computer Sciences at Northeastern as an Assistant Professor in Fall 2022. He obtained his PhD from the University of Maryland. Soheil's research focus is on the theoretical foundations of big data algorithms. He studies algorithms in a variety of large-scale computation settings, including massively parallel computation (MapReduce), graph sparsification, streaming algorithms, and dynamic algorithms.

ABSTRACT
During the past two decades we have seen an unprecedented increase in the amount of data that is being generated from numerous internet-scale applications. As hundreds of millions to billions of users interact with these applications, there is a continuous flow of interaction or log data that is collected by internet companies hosting these applications. Before this data can be subject to modeling and analysis, it becomes necessary to obtain summary statistics such as the cardinality of unique visitors, frequency counts of users from different states or countries, and in general finding the quantile and median information from the dataset. Efficient algorithms exist for computing the exact information over the data. Unfortunately, these algorithms require a considerable amount of time, scanning the data multiple times, or require additional storage that is linear to the size of the dataset itself. Approximation methods, with guaranteed error bounds, developed in the context of streaming data are extremely effective to extract useful and relatively accurate knowledge from big data. In this talk, we will start with a review of one such approximation technique for estimating frequency counts over streaming data where only insert operations are permitted. We will then present refinement of this technique in the context of large datasets that need to handle both insert and delete operations. We will conclude the presentation with an application of the frequency counting technique for an innovative internet caching architecture in which a small front-end caching at the client side significantly mitigates the load imbalance problem at the back-end cache servers. The meta-objective of this talk is to demonstrate the usefulness of approximation techniques in the context of real applications.

SPEAKER BIOGRAPHY
Divyakant Agrawal is a Professor of Computer Science at the University of California at Santa Barbara. His research expertise is in the areas of distributed systems and databases. He is a fellow of the AAAS, ACM, and IEEE. He has also had visiting appointments at IBM Almaden Research Labs, NEC Research, ASK.com, Google, Qatar Computing Research Institute, and National University of Singapore over the course of his professional career.

TITLE
Matrix query language

ABSTRACT
Imagine that instead of the relational data model and relational algebra,
data would be stored as matrices and queries would consist of linear algebra
operators. How would such a matrix query language look like? What can we
do and cannot do in such a language? And how does it relate to query languages
based on first-order logic or the relational algebra? How do these languages
connect to arithmetic circuits, which are often said to "capture" linear algebra?
In this talk we answer some of these questions.

BIO
Floris Geerts holds a Professor position at the University of Antwerp, Belgium. Before that, he was a senior research fellow in the database group at the University of Edinburgh, and held a postdoc position in the data mining group at the University of Helsinki. He received his PhD in 2001 at the University of Hasselt, Belgium. His research interests include the theory and practice of databases, the study of data quality, and more recently, the interaction between linear algebra, relational databases and graph neural networks.

PAPERS

[1] Floris Geerts, Thomas Muñoz, Cristian Riveros, Jan Van den Bussche, Domagoj Vrgoc: Matrix Query Languages. SIGMOD Rec. 50(3): 6-19 (2021). https://sigmodrecord.org/?smd_process_download=1&download_id=%2011461.

[2] Floris Geerts, Thomas Muñoz, Cristian Riveros, Domagoj Vrgoc: Expressive Power of Linear Algebra Query Languages. PODS 2021: 342-354. https://dl.acm.org/doi/10.1145/3452021.3458314.

[3] Robert Brijder, Floris Geerts, Jan Van den Bussche, Timmy Weerwag: On the Expressive Power of Query Languages for Matrices. ACM Trans. Database Syst. 44(4): 15:1-15:31 (2019). https://dl.acm.org/doi/10.1145/3331445.

[4] Floris Geerts: On the Expressive Power of Linear Algebra on Graphs. Theory Comput. Syst. 65(1): 179-239 (2021). https://drops.dagstuhl.de/opus/volltexte/2019/10309/pdf/LIPIcs-ICDT-2019-7.pdf.

ABSTRACT
Graph neural networks (GNNs) are commonly used to compute graph/vertex
embeddings. The capabilities of GNNs are often measured in terms of their distinguishing
power, that is, their ability to distinguish graphs/vertices based on the computed embeddings.
In this talk we view GNNs as queries in a matrix query language, dubbed "Tensor Language".
We show that by using this language, we can obtain insights on the distinguishing power
of GNNs and that it provides an easy toolbox for GNN designers to analyze their own GNNs.

BIO
Floris Geerts holds a Professor position at the University of Antwerp, Belgium. Before that, he was a senior research fellow in the database group at the University of Edinburgh, and held a postdoc position in the data mining group at the University of Helsinki. He received his PhD in 2001 at the University of Hasselt, Belgium. His research interests include the theory and practice of databases, the study of data quality, and more recently, the interaction between linear algebra, relational databases and graph neural networks. 

PAPERS
[1] Floris Geerts, Juan L Reutter: Expressiveness and Approximation Properties of Graph Neural Networks, ICLR 2022.
[2] Martin Grohe: The Logic of Graph Neural Networks. LICS 2021: 1-17.
[3] Martin Grohe: Word2vec, node2vec, graph2vec, X2vec: Towards a Theory of Vector Embeddings of Structured Data. PODS 2020: 1-16.

ABSTRACT
Log-structured merge (LSM) trees are becoming the de facto standard for write-intensive storage layers for both production NoSQL data stores and relational systems. As LSM-based systems are used by various applications and deployed in shared infrastructure (e.g., public or private cloud), they are tasked to support a number of requirements varying from performance, to cost, and privacy, being robust to external requirements and the inherent workload unpredictability. At the heart of any LSM-based engine, we have the background re-organization mechanism (or compaction), the behavior of which affects essentially every aspect of the LSM-tree including write amplification, write throughput, point and range lookup performance, space amplification, and delete performance. In this presentation, we will first introduce in detail the design space of LSM compactions and discuss their tradeoffs, we will then take a deep dive on a new family of delete-aware compactions. We will define as a design goal the delete persistence latency, and discuss how to bound it. Finally, we will discuss the importance of tuning in the presence of uncertainty and present a sneak-peek of a new methodology for near-optimal LSM tuning in the presence of uncertainty of the expected workload vs. the observed one.

BIO
Manos Athanassoulis is an Assistant Professor of Computer Science at Boston University, Director and Founder of the BU Data-intensive Systems and Computing Laboratory and co-director of the BU Massive Data Algorithms and Systems Group. His research is in the area of data management focusing on building data systems that efficiently exploit modern hardware (computing units, storage, and memories), are deployed in the cloud, and can adapt to the workload both at setup time and, dynamically, at runtime. Before joining Boston University, Manos was a postdoc at Harvard University, earlier he obtained his PhD from EPFL, Switzerland, and spent one summer at IBM Research, Watson. Manos’ work has been recognized by awards like “Best of SIGMOD” in 2016, “Best of VLDB” in 2010 and 2017, and “Most Reproducible Paper” at SIGMOD in 2017, and has been supported by NSF and industry funds including two Red Hat Incubation Awards, a Facebook Faculty Research Award and gifts from Meta, Cisco, and Red Hat.

RELEVANT PAPERS (especially 1 & 3)
1. Constructing and Analyzing the LSM Compaction Design Space
Proceedings of the VLDB Endowment, Vol. 14(11), 2021
Subhadeep Sarkar, Dimitris Staratzis, Zichen Zhu, Manos Athanassoulis
2. Reducing Bloom Filter CPU Overhead in LSM-Trees on Modern Storage Devices
Proceedings of the International Workshop on Data Management on New Hardware (DaMoN), 2021
Zichen Zhu, Ju Hyoung Mun, Aneesh Raman, Manos Athanassoulis
3. Lethe: A Tunable Delete-Aware LSM Engine
Proceedings of the ACM SIGMOD International Conference on Management of Data, 2020
Subhadeep Sarkar, Tarikul Islam Papon, Dimitris Staratzis, Manos Athanassoulis
4. And a paper under review: https://arxiv.org/abs/2110.13801

ABSTRACT
Organizations often seek to extend their data by integration with available datasets originating from external sources. While there are many tools that recommend how to perform the integration for given datasets, the selection of what datasets to integrate is often challenging in itself. First, the relevant candidates must be efficiently identified among irrelevant ones. Next, relevant datasets need to be evaluated according to issues such as low quality or poor matching to the target data and schema. Last, jointly integrating multiple datasets may have significant benefits such as increasing completeness and information gain, but may also greatly complicate the task due to dependencies in the integration process.
To assist administrators in this task, we quantify to what extent an integration of multiple datasets is valuable as an extension of an initial dataset and formalize the computational problem of finding the most valuable subset to integrate by this measure. We formally analyze the problem, showing that it is NP-hard; we nevertheless introduce heuristic efficient algorithms, which our experiments show to be near-optimal in practice and highly effective in finding the most valuable integration.

SHORT BIO
Yael Amsterdamer is a Senior Lecturer at the Department of Computer Science, Bar-Ilan University (Ramat-Gan, Israel), and the head of the Data Management Lab. She received her Ph.D. in Computer Science from Tel-Aviv University, and has been a visiting Scholar at the University of Pennsylvania, Philadelphia, PA and jointly at Télécom Paris and INRIA institute (Paris, France). Her research is in the field of data management spanning topics such as crowd-powered data management, provenance and interactive summarization. She has won competitive grants such as the Israeli Science Foundation and Ministry of Science and regularly serves in program committees of top conferences.

RELEVANT PAPER
https://dl.acm.org/doi/10.1145/3459637.3482322

TITLE
Loch Data and Other Monsters: on Creating Data Ecosystems, the Intelligent Way

ABSTRACT
A data ecosystem offers an alliance-driven infrastructure that enables the interaction of different stakeholders and the resolution of interoperability issues among shared data. Despite years of research in data governance and management, trustability is still affected by the absence of transparent and traceable data-driven pipelines. Data integration is the main facilitator of such data-driven pipelines and matching is a task at the heart of any data integration process, aimed at identifying correspondences among data elements. Matching problems were traditionally performed in a semi-automatic manner, with correspondences being generated by matching algorithms and outcomes subsequently validated by human experts. Human-in-the-loop data integration has been recently challenged by the introduction of big data and recent studies have analyzed obstacles to effective human matching and validation. In this talk, we focus on the tension between human and machine matching. We propose a novel data ecosystem architecture that relies on both human knowledge and machine learning and offer a concrete algorithmic solution for effective data integration within this architecture. In particular, we shall present the limitations of human matching and offer a method for learning to characterize reliable and valuable matching experts.

SHORT BIO
Avigdor Gal is The Benjamin and Florence Free Chaired Professor of data science at the Technion - Israel Institute of Technology, where he heads the Big Data Integration laboratory. He specializes in various aspects of data management and mining with about 150 publications in leading journals, books, and conference proceedings. In the current age of big data, his research is focused on developing novel models and algorithms for data integration. In the past he gave keynotes and tutorials in leading conferences in the areas of data and process management. Avigdor Gal is a recipient of the prestigious Yannai award for excellence in academic education, and multiple best paper and test-of-time awards.

Abstract:

The proliferation of inexpensive high-quality cameras coupled
with recent advances in machine learning and computer vision have enabled new applications on video data.
This in turn has renewed interest in video data management systems (VDMSs).
In this talk, we explore how to build a modern data management system
for video data. We focus, in particular, on the storage manager
and present several techniques to store video data in a way that accelerates
queries over that data. We then move up the stack and discuss different
types of data models that can be exposed to applications. Finally, we
discuss the additional challenges of the end-to-end video
analytics pipeline and how a VDMS can support applications throughout
that pipeline.

Bio:

Magdalena Balazinska is Professor and Director of the Paul G. Allen School of Computer Science & Engineering at the University of Washington. Magdalena's research interests are in the field of database management systems. Her current research focuses on data management for data science, big data systems, cloud computing, and image and video analytics. Prior to her leadership of the Allen School, Magdalena was the Director of the eScience Institute, the Associate Vice Provost for Data Science, and the Director of the Advanced Data Science PhD Option. She also served as Co-Editor-in-Chief for Volume 13 of the Proceedings of the Very Large Data Bases Endowment (PVLDB) journal and as PC co-chair for the corresponding VLDB'20 conference. Magdalena is an ACM Fellow. She holds a Ph.D. from the Massachusetts Institute of Technology (2006). Shortly after her arrival at the University of Washington, she was named a Microsoft Research New Faculty Fellow (2007). Magdalena received the inaugural VLDB Women in Database Research Award (2016) for her work on scalable distributed data systems. She also received an ACM SIGMOD Test-of-Time Award (2017) for her work on fault-tolerant distributed stream processing and a 10-year most influential paper award (2010) from her earlier work on reengineering software clones.

TITLE
New ideas and old tasks for reasoning and data management

ABSTRACT
Reasoning is relevant to many aspects of data management, including schema design, data integration, and query optimization. In this talk I give a quick tour of a few major innovative techniques developed in the last decades for applying reasoning to data management. I'll then discuss some of the roadblocks that have been encountered in applying these tools, and mention -- briefly -- some of our current projects to overcome these issues. I'll touch upon work on extending reasoning-based query optimization to complex data types, applying data management-for-reasoning to reasoning-for-data management, mapping rule learning, and applying RL-guided theorem-proving to query optimization.

RELATED PAPERS
1. M. Benedikt. How Can Reasoners Simplify Database Querying (And Why Haven't They Done It Yet)? PODS 2018.
https://doi.org/10.1145/3196959.3196989
2. M. Benedikt, P. Pradic. Generating collection transformations from proofs. POPL 2021.
https://dl.acm.org/doi/10.1145/3434295

BIO
Michael Benedikt is Professor of Computer Science at Oxford University. Prior to that he was distinguished member of technical staff at Bell Laboratories.

ABSTRACT:
The public cloud emerged a decade ago, yet distributed systems are still programmed using models from sequential computing. All the traditional challenges of distributed programming and data are still present in the cloud, only they are now faced by the general population of software developers. Added to these challenges are new desires for "serverless" computing, including consumption-based pricing and autoscaling.
 
This talk will highlight principles for cloud programming that I have explored with colleagues over the past decade, including the CALM Theorem and languages like Dedalus and Bloom that encourage monotonic coordination-free consistency via logic and lattices. The Anna "any-scale" KVS will be presented as a petri dish for the potential of these ideas and many remaining challenges.

Time permitting, I will conclude by overviewing new work in the Hydro project, which is aimed at bringing research ideas to programmers in an practical, evolutionary fashion. Key to our approach is a separation of distributed programs into a PACT of four facets: Program semantics, Availablity, Consistency and Trust. We propose to migrate developers gradually to PACT programming by lifting familiar code into our more declarative level of abstraction. This agenda raises challenges across multiple areas including language design, query optimization, transactions, distributed consistency, compilers and program synthesis.

BIO:
Joe Hellerstein is the Jim Gray Professor of Computer Science at the University of California, Berkeley, and currently on leave as a Faculty Fellow at Sutter Hill Ventures. His research focuses on data-centric systems and the way they drive computing. In addition to his academic work, Hellerstein has been involved in a number of startup companies including Trifacta, which brought academic research on data wrangling to market.

Relevant papers:

• CALM Theorem: Keeping CALM: When Distributed Consistency is Easy (CACM 09/20) (https://cacm.acm.org/magazines/2020/9/246941-keeping-calm/fulltext)
• Future Research Directions: New Directions in Cloud Programming (CIDR 2021) (http://cidrdb.org/cidr2021/papers/cidr2021_paper16.pdf)
• Declarative distributed programming:
  • Dedalus: Datalog in Time and Space  (Datalog 2.0 2010) (http://www.neilconway.org/docs/dedalus_dl2.pdf)
  • Bloom: Logic and Lattices for Distributed Programming (SoCC 2012) (https://web.archive.org/web/20130108035944id_/http://db.cs.berkeley.edu/papers/socc12-blooml.pdf)

Abstract:
In the first part of the talk, I will revisit various notions of decompositions and widths and recapitulate some of their crucial properties. In particular, we thus mention:
*) hypertree-width (hw)
*) generalized hypertree-width (ghw)
*) fractional hypertree-width (fhw)

Time permitting, I will then present some recent results on the task of actually computing such decompositions:
*) NP-hardness of checking if fhw or ghw is at most 2
*) Tractable cases of ghw and fhw computation
*) Parallel approach to computing ghw and hw

Curriculum Vitae:
I got a master's degree from the University of Innsbruck (Mag.rer.nat. in Mathematics, 1991), another master's degree from the University of London (MSc in Mathematical Computation, 1992, at the QMW College), and a PhD from the TU Wien (Dr.techn. in Computer Science, 2000). In May 2001, I received my "habilitation" in Theoretical Computer Science.

From 1992 to 2005, I worked in the Program and Systems Engineering Department (PSE) of the Siemens AG Österreich. Since July 2005, I have been a professor in the Databases and Artificial Intelligence Group (DBAI) of the TU Wien.

Relevant publications:

Wolfgang Fischl, Georg Gottlob, Reinhard Pichler:
General and Fractional Hypertree Decompositions: Hard and Easy Cases. PODS 2018: 17-32.
(full version in J. ACM, 2021)

Wolfgang Fischl, Georg Gottlob, Davide Mario Longo, Reinhard Pichler: HyperBench: A Benchmark and Tool for Hypergraphs and Empirical Findings. PODS 2019: 464-480
(full version in ACM J. Exp. Alg., 2021)

Georg Gottlob, Matthias Lanzinger, Reinhard Pichler, Igor Razgon: Fractional Covers of Hypergraphs with Bounded Multi-Intersection. MFCS 2020: 41:1-41:14

Hubie Chen, Georg Gottlob, Matthias Lanzinger, Reinhard Pichler: Semantic Width and the Fixed-Parameter Tractability of Constraint Satisfaction Problems. IJCAI 2020: 1726-1733

Georg Gottlob, Cem Okulmus, Reinhard Pichler:
Fast and Parallel Decomposition of Constraint Satisfaction Problems. IJCAI 2020: 1155-1162.

Georg Gottlob, Matthias Lanzinger, Cem Okulmus, Reinhard Pichler: Fast Parallel Hypertree Decompositions in Logarithmic Recursion Depth. CoRR abs/2104.13793 (2021)
To appear at PODS 2022

Abstract:
Pre-training/fine-tuning paradigms have transformed the natural language processing field. For tabular data-based tasks, however, their potential has been far less explored. In this talk, we will discuss our recent efforts: (1) TURL, a pre-training/fine-tuning paradigm on relational Web tables, which benefits a wide range of tasks for table understanding (e.g., relation extraction, entity linking). (2) StruG, a weakly supervised Structure-Grounded pretraining framework for text-to-SQL, which effectively learns to capture the text-table alignment essential for the task. (3) ReasonBERT, a pre-training method that augments language models for multi-step reasoning over hybrid contexts (textual and tabular). Among them, we will cover TURL in greater detail. Finally, we will conclude the talk with some of the open questions in the space.

Bio:
Huan Sun is an assistant professor in the Department of Computer Science and Engineering at the Ohio State University. Before joining OSU, she was a visiting scientist at the University of Washington (01-06/2016), received a Ph.D. in Computer Science from University of California, Santa Barbara (2015) and a B.S. in EEIS from the University of Science and Technology of China (2010). Her research interests lie in natural language processing, data mining and management, and artificial intelligence, with emphasis on building various kinds of natural language interfaces, task-oriented dialogue and conversational AI systems. Huan received the 2022 SIGMOD Research Highlight Award, 2021 BIBM Best Paper Award, Google Research Scholar Award (2022), NSF CAREER Award (2020), OSU Lumley Research Award (2020), SIGKDD Ph.D. Dissertation Runner-Up Award (2016), among others.

Relevant papers:
Turl: https://arxiv.org/abs/2006.14806 Please feel free to check out more papers on learning neural representations for tables mentioned in this tutorial. Unfortunately, we won’t be able to cover them in this talk.
StruG: https://arxiv.org/abs/2010.12773
ReasonBert: https://arxiv.org/abs/2109.04912

ABSTRACT
Data analytics in the cloud has become an integral part of enterprise businesses. Big data analytics systems, however, still lack the ability to take task objectives such as user performance goals and budgetary constraints and automatically configure an analytical job to achieve these objectives. This talk presents UDAO, a Unified Data Analytics Optimizer, that can automatically determine a cluster configuration with a suitable number of cores as well as other system parameters that best meet the task objectives. At a core of our work is a principled multi-objective optimization (MOO) approach that computes a Pareto optimal set of configurations to reveal tradeoffs between different objectives, recommends a new cluster configuration that best explores such tradeoffs, and employs novel optimizations to enable such recommendations within a second. Such optimization is further enabled by a Deep Learning-based modeling approach that can learn a model for each user objective as complex as necessary for the underlying computing environment. Evaluation using production workloads shows that our optimizer could reduce 36-67% latency and, at the same time, 37-75% cost in an existing cloud analytics system, while running under 200 msec.

BIO
Yanlei Diao is Professor of Computer Science at Ecole Polytechnique, France and the University of Massachusetts Amherst, USA. She recently joined Amazon as an Amazon Scholar. Her research interests lie in big data analytics and scalable intelligent information systems, with a focus on optimization in cloud analytics, data stream analytics, explanation discovery, interactive data exploration, and uncertain data management. She received her PhD in Computer Science from the University of California, Berkeley in 2005.

Prof. Diao was a recipient of the 2016 ERC Consolidator Award, 2013 CRA-W Borg Early Career Award (one female computer scientist selected each year for outstanding contributions), IBM Scalable Innovation Faculty Award, and NSF Career Award. She has given keynote speeches at the ACM DEBS Conference, the ExploreDB workshop, and the Distinguished Lecture Series at the IBM Almaden Research Center, the University of Texas at Austin and Technische Universitaet Darmstadt. She has served as Editor-in-Chief of the ACM SIGMOD Record, Associate Editor of ACM TODS, Chair of the ACM SIGMOD Research Highlight Award Committee, and member of the SIGMOD and PVLDB Executive Committees. She was PC Co-Chair of IEEE ICDE 2017 and ACM SoCC 2016, and served on the organizing committees of SIGMOD, PVLDB, and CIDR, as well as on the program committees of many international conferences and workshops.