ICLP 2021

Neural language models, which can be pretrained on very large corpora, turn out to "know" a lot about the world, in the sense that they can be trained to answer questions surprisingly reliably. However, "language models as knowledge graphs" have many disadvantages: for example, they cannot be easily updated when information changes. I will describe recent work in my team and elsewhere on incorporating symbolic knowledge into language models and question-answering systems, and also comment on some of the remaining challenges associated with integrating symbolic reasoning and neural NLP.

Bio:

William Cohen is a Principal Scientist at Google, and is based in Google's Pittsburgh office. He received his bachelor's degree in Computer Science from Duke University in 1984, and a PhD in Computer Science from Rutgers University in 1990. From 1990 to 2000 Dr. Cohen worked at AT&T Bell Labs and later AT&T Labs-Research, and from April 2000 to May 2002 Dr. Cohen worked at Whizbang Labs, a company specializing in extracting information from the web. From 2002 to 2018, Dr. Cohen worked at Carnegie Mellon University in the Machine Learning Department, with a joint appointment in the Language Technology Institute, as an Associate Research Professor, a Research Professor, and a Professor. Dr. Cohen also was the Director of the Undergraduate Minor in Machine Learning at CMU and co-Director of the Master of Science in ML Program. Dr. Cohen is a past president of the International Machine Learning Society, and was General Chair for the 2008 International Machine Learning Conference, Program Co-Chair of the 2006 International Machine Learning Conference; and Co-Chair of the 1994 International Machine Learning Conference. Dr. Cohen was also the co-Chair for the 3rd Int'l AAAI Conference on Weblogs and Social Media, and co-Program Chair for the 4rd Int'l AAAI Conference on Weblogs and Social Media. He is a AAAI Fellow, and was a winner of the 2008 SIGMOD "Test of Time" Award for the most influential SIGMOD paper of 1998, and the 2014 SIGIR "Test of Time" Award for the most influential SIGIR paper of 2002-2004.

Dr. Cohen's research interests include information integration and machine learning, particularly information extraction, text categorization and learning from large datasets. He has a long-standing interest in statistical relational learning and learning models, or learning from data, that display non-trivial structure. He holds seven patents related to learning, discovery, information retrieval, and data integration, and is the author of more than 300 publications.

John Hooker, Carnegie Mellon University --- Logic-Based Formulation of Ethical Principles (slides)

Quantified modal logic provides a useful framework for precise formulation of ethical principles. This talk begins by briefly summarizing how generalization, utilitarian, and autonomy principles can be deontologically derived from the logical structure of action. It then shows how these principles can be expressed in a logical idiom suitable for use in a rule-based AI system. In particular, it indicates how quantified modal logic can enable value alignment in machine learning to integrate ethical reasoning with empirical observation. Portions of the talk are based on joint work with Thomas Donaldson and Tae Wan Kim.

Bio:

John Hooker is professor of operations research and Holleran Professor of Business Ethics and Social Responsibility at Carnegie Mellon University. His primary research interests are logic-based methods for optimization, constraint programming, mathematical analysis of distributive justice, ethics of AI and machine learning, business ethics, deontological foundations for ethics, cross-cultural management, and music theory. He is an INFORMS Fellow and recipient of the INFORMS Computing Society Award, the INFORMS Khachiyan Award for lifetime achievements in optimization, and the Research Excellence Award of the Association for Constraint Programming.

Phokion G. Kolaitis, UC Santa Cruz and IBM Research --- Structural Characterizations of Rule-Based Languages (slides)

Since the early days of the relational data model, rule-based languages have been used to express semantic restrictions that the data of interest ought to obey. In the past two decades, rule-based languages have also been used as specification languages in data exchange, in data integration, and in ontology-mediated data access. Tuple-generating dependencies (tgds) form one of the most widely used and extensively studied such rule-based languages. Tgds are also known as existential rules; they include Datalog rules as a special case and possess numerous desirable structural properties. The aim of this talk is to present an overview of results, both old and new, that characterize finite sets of tgds in terms of their structural properties. These results include characterizations of finite sets of arbitrary tgds, as well as characterizations of restricted types of tgds, including full tgds, linear tgds, guarded tgds, and source-to-target tgds.

Bio:

Phokion Kolaitis is a Distinguished Research Professor of Computer Science and Engineering at the University of California Santa Cruz and a Principal Research Staff Member at the IBM Almaden Research Center. His research interests include principles of database systems, logic in computer science, and computational complexity.

Kolaitis is a Fellow of the American Association for the Advancement of Science (AAAS), a Fellow of the Association for Computing Machinery (ACM), a Foreign Member of Academia Europaea, and a Foreign Member of the Finnish Academy of Science and Letters. He has held a Guggenheim Fellowship and is a co-winner of the 2020 ACM SIGLOG Alonzo Church Award, a co-winner of both the 2008 and the 2014 ACM PODS Alberto O. Mendelzon Test-of-Time Award, and a co-winner of the 2013 International Conference on Database Theory Test-of-Time Award.

Stuart Russell, University of California, Berkeley --- Combining probability and first-order logic (slides)

Probabilistic programming languages (PPLs) are expressive formal languages for writing probability models. One branch derives expressive power from traditional programming languages; the other from first-order logic. This talk covers Bayesian logic (BLOG), a language that facilitates specification of probability distributions over first-order structures. I will describe the language mainly through examples and cover its application to monitoring the Comprehensive Nuclear-Test-Ban Treaty. PPLs have many advantages over deep learning systems and may offer an alternative route to the construction of general-purpose AI systems.

Bio:

Stuart Russell is a Professor of Computer Science at the University of California at Berkeley, holder of the Smith-Zadeh Chair in Engineering, and Director of the Center for Human-Compatible AI. He is a recipient of the IJCAI Computers and Thought Award and held the Chaire Blaise Pascal in Paris. In 2021 he received the OBE from Her Majesty Queen Elizabeth. He is an Honorary Fellow of Wadham College, Oxford, an Andrew Carnegie Fellow, and a Fellow of the American Association for Artificial Intelligence, the Association for Computing Machinery, and the American Association for the Advancement of Science. His book "Artificial Intelligence: A Modern Approach" (with Peter Norvig) is the standard text in AI, used in 1500 universities in 135 countries. His research covers a wide range of topics in artificial intelligence, with a current emphasis on the long-term future of artificial intelligence and its relation to humanity. He has developed a new global seismic monitoring system for the nuclear-test-ban treaty and is currently working to ban lethal autonomous weapons.

Jeff Ullman, Stanford University and Gradiance Corp --- Why Datalog? (slides)

(This became part of a special session on Datalog---see immediately below---that included also invited talks by David Maier and Moshe Vardi, chaired by David S. Warren.)

Special Session on Datalog: Invited Talks and Panel (abstracts, slides, video, photos)

Video

Photos

Datalog Perspectives

Chair: David S. Warren, Stony Brook University and XSB, Inc.

Datalog was identified as a language and named about 35 years ago now. It has gone through phases of popularity and unpopularity. It has spawned (at least) two distinct, quite different, intuitive and formal semantics. It has stimulated much research. In this session we hear from several researchers, who did early and foundational research on Datalog, to get their perspectives on Datalog: what it is, where it's been, and where it's going.

Discussion Questions:

What is the role (if any) for non-stratified Datalog specifications?

Thoughts on the multiple Datalog semantics? e.g. Well-founded and Stable?

What are the outstanding (remaining) formal and practical research questions regarding Datalog?

What's the difference between Data representation and Knowledge representation? What is Datalog's role?

Jeff Ullman, Stanford University and Gradiance Corp --- Why Datalog? (slides)

Datalog was far from the first brand of logic to be thought of as a database query language. Other approaches were far more powerful. But its weakness was in fact its strength. Its advantage is that a logic as limited as Datalog makes optimization feasible, and for any declarative query language, the ability to select an efficient implementation from among the many possible ways to execute a query is essential.

Bio:

Jeff Ullman is the Stanford W. Ascherman Professor of Engineering (Emeritus) in the Department of Computer Science at Stanford and CEO of Gradiance Corp. He received the B.S. degree from Columbia University in 1963 and the PhD from Princeton in 1966. Prior to his appointment at Stanford in 1979, he was a member of the technical staff of Bell Laboratories from 1966-1969, and on the faculty of Princeton University between 1969 and 1979. From 1990-1994, he was chair of the Stanford Computer Science Department. Ullman was elected to the National Academy of Engineering in 1989, the American Academy of Arts and Sciences in 2012, the National Academy of Sciences in 2020, and has held Guggenheim and Einstein Fellowships. He has received the Sigmod Contributions Award (1996), the ACM Karl V. Karlstrom Outstanding Educator Award (1998), the Knuth Prize (2000), the Sigmod E. F. Codd Innovations award (2006), the IEEE von Neumann medal (2010), the NEC C&C Foundation Prize (2017), and the ACM A.M. Turing Award (2020). He is the author of 16 books, including books on database systems, data mining, compilers, automata theory, and algorithms.

David Maier, Portland State University --- The Downs and Ups of Datalog (slides)

After a burst of interest in the late 1980s and early 1990s, activity around Datalog and other logic languages declined. However, Datalog has garnered renewed attention in the last decade, often for uses beyond database querying. I will explore possible reasons for the drop-off in interest, cover a range of current application areas for Datalog, and speculate on the reasons for resurgence.

Bio:

DAVID MAIER is Maseeh Professor of Emerging Technologies at Portland State University. Prior to his current position, he was on the faculty at the State University of New York, Stony Brook; Oregon Graduate Institute; and Oregon Health and Science University. He has spent extended visits with Inria, the University of Wisconsin–Madison, Microsoft Research, and the National University of Singapore. He is the author of books on relational databases, logic programming, and object-oriented databases, as well as papers on database theory, object-oriented technology, scientific databases, and dataspace management. He received an NSF Young Investigator Award in 1984 and was awarded the 1997 ACM SIGMOD’s Innovations Award for his contributions in objects and databases. He is also an ACM Fellow and IEEE Senior Member. He holds a dual B.A. in mathematics and in computer science from the University of Oregon (Honors College, 1974) and a Ph.D. in electrical engineering and computer science from Princeton University (1978). He is credited with coining the name "Datalog" for the function-free subset of Prolog.

Moshe Vardi, Rice University --- Datalog++ and Datalog-- (slides)

One argument in favor of Datalog is that it constitutes a sweet point in terms of expressiveness and feasibility of evaluation and optimization. I will argue that instead of thinking of Datalog as a single query language, we should consider it as a family of query languages. For some applications, such as universal relations and data exchange, we need to enhance Datalog's expressive power, hence Datalog++. For other applications, where we need query equivalence to be decidable, we need to limit Datalog's expressiveness, hence Datalog--.

Bio:

Moshe Y. Vardi is a University Professor and the George Distinguished Service Professor in Computational Engineering at Rice University. He is the recipient of three IBM Outstanding Innovation Awards, the ACM SIGACT Goedel Prize, the ACM Kanellakis Award, the ACM SIGMOD Codd Award, the Blaise Pascal Medal, the IEEE Computer Society Goode Award, the EATCS Distinguished Achievements Award, the Southeastern Universities Research Association's Distinguished Scientist Award, and the ACM SIGLOG Church Award. He is the author and co-author of over 650 papers, as well as two books: Reasoning about Knowledge and Finite Model Theory and Its Applications. He is a Fellow of the American Association for the Advancement of Science, the American Mathematical Society the Association for Computing Machinery, the American Association for Artificial Intelligence, the European Association for Theoretical Computer Science, the Institute for Electrical and Electronic Engineers, and the Society for Industrial and Applied Mathematics. He is a member of the US National Academy of Engineering and National Academy of Science, the American Academy of Arts and Science, the European Academy of Science, and Academia Europaea. He holds seven honorary doctorates. He is currently a Senior Editor of the Communications of the ACM, after having served for a decade as Editor-in-Chief.

Invited Panel (abstract, slides, video)

Video

No Logic is an Island: Internal and External Integration of Logic Programming Paradigms

Chair: Joost Vennekens, KU Leuven

Panelists: Marc Denecker (slides), Manuel Hermenegildo (slides), Francesco Ricca (slides), Theresa Swift (slides), Jan Wielemaker (slides)

Over the course of its rich history, the area of Logic Programming has spawned many different logics, systems and paradigms. In this panel, we ask two questions:

(1) How can different subareas/subformalisms of Logic Programming (Prolog, Datalog, ASP, CSP, (co-)induction, ...) be integrated with each other, and is there something to be gained by such an integration?

(2) How can key ideas from Logic Programming be integrated with other paradigms (imperative/functional programming, description logic and semantic web, databases, ...) , and is it desirable to do so?

We ask these questions with both practical and theoretical goals in mind:

- In order to build real-life applications, there may be a practical need to integrate different tools and technologies.

- In order to have an enduring scientific impact in the general domain of Computer Science, it may be helpful for key ideas from logic programming to be integrated with other paradigms.

In this panel, we reflect on achievements so far and useful directions for future efforts on these topics.

Test-of-Time Award Talks (slides)

20-Year Award: Ultimate Well-Founded and Stable Semantics for Logic Programs with Aggregates (slides)
Marc Denecker, Nikolay Pelov, and Maurice Bruynooghe

10-Year Award: The PITA System: Tabling and Answer Subsumption for Reasoning under Uncertainty (slides)
Fabrizio Riguzzi and Theresa Swift

Page updated

Report abuse