Greeting/Introduction

Speaker Biography

Dr. Daniel Dongyuel Lee is currently the Tisch University Professor in Electrical and Computer Engineering at Cornell Tech and Executive Vice President for Samsung Research. He previously was the UPS Foundation Chair Professor in the School of Engineering and Applied Science at the University of Pennsylvania. He received his B.A. summa cum laude in Physics from Harvard University and his Ph.D. in Condensed Matter Physics from the Massachusetts Institute of Technology. He was also a researcher at Bell Labs in the Theoretical Physics and Biological Computation departments. He is a Fellow of the IEEE and AAAI and has received the National Science Foundation CAREER award and the Lindback award for distinguished teaching. He was also a fellow of the Hebrew University Institute of Advanced Studies in Jerusalem, an affiliate of the Korea Advanced Institute of Science and Technology, and organized the US-Japan National Academy of Engineering Frontiers of Engineering symposium and Neural Information Processing Systems (NeurIPS) conference. His group focuses on understanding general computational principles in biological systems and on applying that knowledge to build autonomous systems.

Towards a Foundation for AGI

Large pre-trained language models like GPT-3 and PaLM have generated enormous enthusiasm, and are capable of producing remarkably fluent language. But they have also been criticized on many grounds, and described as "stochastic parrots." Are they adequate as a basis for general intelligence, and if not, what would a better foundation for general intelligence look like?

Speaker Biography

Gary Marcus is an American scientist, author, and entrepreneur who is a professor in the Department of Psychology at New York University and was the founder and CEO of Geometric Intelligence, a machine learning company later acquired by Uber. His books include Guitar Zero, which appeared on the New York Times Best Seller list, and Kluge: The Haphazard Construction of the Human Mind, a New York Times Editors' Choice. With Jeremy Freeman, he was co-editor of The Future of the Brain: Essays by the World's Leading Neuroscientists.

Subgraph Aggregation Networks

Message-passing neural networks (MPNNs) are the leading architecture for deep learning on graph-structured data, in large part due to their simplicity and scalability. Unfortunately, it was shown that these architectures are limited in their expressive power. In order to gain more expressive power, a recent trend applies message-passing neural networks to subgraphs of the original graph. In this talk, I will present a representative framework of this family of methods, called Equivariant Subgraph Aggregation Networks (ESAN). The main idea behind ESAN is to represent each graph as a set of subgraphs derived from a predefined policy and to process the set of subgraphs using a suitable equivariant architecture. Our analysis shows that ESAN has favorable theoretical properties and that it performs well in practice. Following this, we will discuss some special properties of popular subgraph selection policies by connecting subgraph GNNs with previous work in equivariant deep learning.

Speaker Biography

Haggai is a Research Scientist at NVIDIA Research and a member of NVIDIA's TLV lab. His main field of interest is machine learning in structured domains. In particular, he works on applying deep learning to sets, graphs, point clouds, and surfaces, usually by leveraging their symmetry structure. He completed his Ph.D. in 2019 at the Weizmann Institute of Science under the supervision of Prof. Yaron Lipman. Haggai will be joining the Faculty of Electrical and Computer Engineering at the Technion as an Assistant Professor in 2023.

Foundations of Symbolic Languages for Model Interpretability

Several queries and scores have recently been proposed to explain individual predictions over ML models. Examples include queries based on “anchors”, which are parts of an instance that are sufficient to justify its classification, and “feature-perturbation” scores such as SHAP. Given the need for flexible, reliable, and easy-to-apply interpretability methods for ML models, we foresee the need for developing declarative languages to naturally specify different explainability queries. We do this in a principled way by rooting such a language in a logic called FOIL, which allows for expressing many simple but important explainability queries, and might serve as a core for more expressive interpretability languages. We study the computational complexity of FOIL queries over two classes of ML models often deemed to be easily interpretable: decision trees and more general decision diagrams. Since the number of possible inputs for an ML model is exponential in its dimension, tractability of the FOIL evaluation problem is delicate but can be achieved by either restricting the structure of the models, or the fragment of FOIL being evaluated. We also study the complexity of basic problems related to model interpretability under probabilistic interpretations.

Speaker Biography

Full Professor at Pontificia Universidad Católica de Chile, where he also acts as Director of the Institute for Mathematical and Computational Engineering. Ph.D. in Computer Science from the University of Toronto, Canada. He is an associate researcher of the Millennium Institute for Foundational Research on Data (IMFD Chile) and the National Center for Artificial Intelligence (CENIA). He is the author of more than 90 technical papers, has chaired ICDT 2019, will be chairing ACM PODS 2022, and is currently a member of the editorial committee of Logical Methods in Computer Science. From 2011 to 2014 he was the editor of the database theory column of SIGMOD Record. His areas of interest are database theory, logic in computer science, and the emerging relationship between these areas and machine learning.

Modeling Symbolic Domains via Compositional and Geometric Representations

In this talk, I will discuss representation learning in symbolic domains and how to use such models for simple reasoning tasks. I will first present compositional models of symbolic knowledge representations such as tensor-product and holographic models, discuss their connections to associative memory, and show that they are able to outperform purely symbolic methods in various deductive reasoning settings. Furthermore, I will discuss how structural properties of symbolic data such as hierarchies and cycles are connected to the geometry of a representation space and how geometric representation learning enables parsimonious models that preserve important semantic properties of the domain. Moreover, I will show how such embeddings can be applied to challenging tasks in NLP and biology. In addition, I will discuss connections of geometric representations to state-of-the-art generative models such Riemmannian continous normalizing flows and Moser flow.

Speaker Biography

Maximilian Nickel is a Research Scientist at Facebook AI Research in New York. Before joining FAIR, he was a postdoctoral fellow at MIT where he was with the Laboratory for Computational and Statistical Learning and the Centre for Brains, Minds, and Machines. In 2013, he received his Ph.D. summa cum laude from the Ludwig Maximilian University Munich. From 2010 to 2013 he worked as a research assistant at Siemens Corporate Technology. His research centers around geometric methods for learning and reasoning with relational knowledge representations and their applications in artificial intelligence and network science.

Knowledge Graph Embeddings: Models, Methods, Future?

Knowledge graph embedding (KGE) models represent the entities and relations of a knowledge graph using dense embeddings. They are simple, can provide good empirical results, and are the subject of a (very!) large number of research papers. In this talk, I'll provide an introduction to KGE models, their training and evaluation methods, and as well as my thoughts about their potential usefulness, limitations, and future. The talk starts with key concepts, then highlights selected contributions (viewed through the lens of what we know now) and variants, and concludes with a critical assessment of the field.

Speaker Biography

Rainer Gemulla is a professor at the University of Mannheim, where he heads the Data Analytics group. His research focuses on methods and systems for scalable data analytics, data mining, and machine learning. Recent examples include work on representation learning for knowledge graphs (including the LibKGE project for reproducible KGE research, and the KGT5 sequence-to-sequence model) and on efficient parameter management for parallel machine learning (including the DistKGE framework for parallel KGE training and the NuPS parameter server). Rainer has published numerous papers in top-ranked international conferences (such as ACL, AAAI, ICDE, ICDM, ICLR, PVLDB, KDD, SIGMOD) and journals (such as TKDD, TODS, VLDB Journal). He has been appointed a Junior Fellow of the German Informatics Society in 2013 and received multiple awards for his work.

Generative Models and the Mismeasure of Artificial Intelligence

Data-driven methods like deep nets have led to enormous advances in computer vision and they typically vastly outperform alternative methods such as generative models with symbolic representations. But it is paradoxical that despite claims of superhuman performance, AI algorithms are far less robust and general-purpose than the human visual system and often fail in surprising ways in real-world conditions. This paradox arises partly because academic performance evaluation mismeasures the performance of AI algorithms when faced with the complexity of the real world. This motivates tougher performance measures including out-of-distribution and out-of-task testing. More radically, we also propose that AI algorithms should be tested by adversarial examiners who search over the vast space of stimuli to determine the weak points of algorithms. We argue that generative methods, which are consistent with studies of human perception, will outperform standard deep nets when faced with these tougher performance measures. As proof of concept, we present compositional generative networks and show that they can perform as well as deep nets using standard benchmarks but outperform them when evaluated with these tougher performance measures.

Speaker Biography

Alan Yuille received his B.A. in mathematics from the University of Cambridge in 1976, and completed his Ph.D. in theoretical physics at Cambridge in 1980 on quantum gravity supervised by Stephen Hawking. He did postdoctoral research on theoretical physics in 1981 at the Physics Department, the University of Texas at Austin, and the Institute for Theoretical Physics, Santa Barbara. In 1982 he changed research fields and became a research scientist at the Artificial Intelligence Laboratory at MIT (1982-1986) specializing in computer vision and theoretical neuroscience. He moved to the Division of Applied Sciences at Harvard University in 1986 as a postdoctoral researcher and then served as a Professor of Engineering from 1988-1996 rising to the level of Associate Professor. From 1995-2002 he worked as a senior scientist at the Smith-Kettlewell Eye Research Institute in San Francisco. From 2002-2016 he was a full professor in the Department of Statistics at UCLA with joint appointments in Psychology, Computer Science, and Psychiatry. He also held a part-time appointment at Korea University as part of the World Class University (WCU) program. In 2016 he became a Bloomberg Distinguished Professor in Cognitive Science and Computer Science at Johns Hopkins University. He has won a Marr prize, a Helmholtz prize, and is a Fellow of IEEE. He has broad research interests in vision, machine learning, cognitive science, neuroscience, and medical image analysis. He has published over 500 research articles, co-authored two books -- Data Fusion for Sensory Information Processing Systems (with J. J. Clark) and Two- and Three-Dimensional Patterns of the Face (with P. W. Hallinan, G. G. Gordon, P. J. Giblin, and D. B.Mumford) – and he also co-edited the book Active Vision (with A. Blake). He has over 75,000 citations and a current h-number of 118.

The Knowledge, Action Cycle

-

Speaker Biography

Abhinav Gupta is an Associate Professor at the Robotics Institute, Carnegie Mellon University and Research Manager at Facebook AI Research (FAIR). His research focuses on scaling up learning by building self-supervised, lifelong, and interactive learning systems. Specifically, he is interested in how self-supervised systems can effectively use data to learn visual representation, common sense, and representation for actions in robots. Abhinav is a recipient of several awards including IAPR 2020 JK Aggarwal Prize, PAMI 2016 Young Researcher Award, ONR Young Investigator Award, Sloan Research Fellowship, Okawa Foundation Grant, Bosch Young Faculty Fellowship, YPO Fellowship, IJCAI Early Career Spotlight, ICRA Best Student Paper award, and the ECCV Best Paper Runner-up Award. His research has also been featured in Newsweek, BBC, Wall Street Journal, Wired, and Slashdot.

Neural network function without learning: How can nature achieve that?

It is reasonable to assume that the human brain acquires symbolic and logical reasoning capabilities not through learning, but through genetically encoded structural properties of neural networks of the human brain. But it has remained an open problem what these structural properties are. An understanding of them would not only provide substantial progress in brain science but also inspire new methods for endowing artificial neural networks with similar "innate" capabilities. Unfortunately, we are still far away from understanding those innate structural properties of neural networks in the brain that provide symbolic reasoning capabilities. But experimental data on generic cortical microcircuits elucidate the structural properties of neural networks in the brain that are likely to be involved. I will show that these structural features that are under genetic control provide a quite powerful "programming language" for inducing specific computational capabilities in neural networks, without a need for synaptic plasticity or other forms of learning. This insight gives rise to a new research program for solving the open problem of structural features of neural networks that are likely to induce symbolic and logical reasoning capabilities.

Speaker Biography

Phd (1974) and Habilitation (1978) in Mathematics at the Ludwig-Maximilians-Universitaet in Munich. From 1979 - 1984 research at MIT, the University of Chicago, and the University of California at Berkeley as a Heisenberg-Fellow of the Deutsche Forschungsgemeinschaft. From 1982 - 1986 Associate Professor and from 1986 - 1993 Professor of Computer Science at the University of Illinois in Chicago. Since 1991 Professor of Computer Science at the Graz University of Technology in Austria, and 1992-2017 Head of the Institut fuer Grundlagen der Informationsverarbeitung (Institute of Theoretical Computer Science) at the Graz University of Technology. Sloan Fellow at the Computational Neurobiology Lab of the Salk Institute (La Jolla, USA) during 1997/98. From 9/2002 - 2/2003 and from 4/2012 - 7/2012 Visiting Professor at the Brain-Mind Institute, EPFL, Lausanne, Switzerland. Since 2005 Adjunct Fellow of the Frankfurt Institute of Advanced Studies (FIAS) From 2008 - 2012 Member of the Board of Governors of the International Neural Network Society. Since 2013 Member of the Academia Europaea. 2018: Co-Organizer of the Special Semester "The Brain and Computation" at the Simons Institute, University of California at Berkeley.

Injecting Logical Constraints into Neural Networks

Integrating low-level perception with high-level reasoning has been a fundamental challenge in AI. Despite the vast success of deep neural networks, it is not clear how knowledge can be conveniently represented and how complex high-level reasoning can be computed in deep neural networks. I will describe two methods in this direction. NeurASP integrates neural networks with Answer Set Programming (ASP), a declarative logic programming. It provides a convenient way to represent structured knowledge and train a neural network better so that a neural network not only learns from implicit correlations from the data but also from the explicit complex semantic constraints expressed by the ASP rules. Next, to have a more efficient computation, I will describe a systematic way to represent logical constraints as a loss function borrowing a method used to train binary neural networks to make a discretizing function meaningfully differentiable.

Speaker Biography

Joohyung Lee is the head of the AI Methods Team at Samsung Research and an associate professor in the School of Computing and Augmented Intelligence at Arizona State University. His research interests include knowledge representation, machine learning, neuro-symbolic AI, question answering, and logic programming.

On Reasoning, Planning, and Incidental Supervision

-

Speaker Biography

Dan Roth is the Eduardo D. Glandt Distinguished Professor at the Department of Computer and Information Science, University of Pennsylvania, lead of NLP Science at AWS AI Labs., and a Fellow of the AAAS, the ACM, AAAI, and the ACL. In 2017 Roth was awarded the John McCarthy Award, the highest award the AI community gives to mid-career AI researchers. Roth was recognized “for major conceptual and theoretical advances in the modeling of natural language understanding, machine learning, and reasoning.” Roth has published broadly in machine learning, natural language processing, knowledge representation and reasoning, and learning theory, and has developed advanced machine learning-based tools for natural language applications that are being used widely. Roth was the Editor-in-Chief of the Journal of Artificial Intelligence Research (JAIR) and a program chair of AAAI, ACL, and CoNLL. Roth has been involved in several startups; most recently he was a co-founder and chief scientist of NexLP, a startup that leverages the latest advances in Natural Language Processing (NLP), Cognitive Analytics, and Machine Learning in legal and compliance domains. NexLP was acquired by Reveal in 2020. Prof. Roth received his B.A Summa cum laude in Mathematics from the Technion, Israel, and his Ph.D. in Computer Science from Harvard University in 1995.

Counterfactual-Based Explanations for Classification Outcomes: Semantics and Reasoning

In this talk, we revisit score-based attribution explanations for outcomes of classification models. Some of them are based on the use of explicit or implicit counterfactual interventions. We will consider two important aspects of this kind of explanations, namely the addition and use of domain semantics, and the use of logical reasoning with counterfactual interventions in this setting, for exploring different scenarios, and also adding domain knowledge via the reasoning system.

Speaker Biography

Leopoldo Bertossi is a Full Professor at the Faculty of Engineering and Sciences, "Universidad Adolfo Ibáñez" (UAI, Santiago, Chile), where he is the Director of the Graduate Programs in Data Science. He is also a Senior Fellow of the UAI. In the next few months, he will be moving as a Professor to the Skema Business School, R&D Lab for Business AI, Montreal, Canada. He is an Emeritus Professor of the School of Computer Science, Carleton University (Ottawa, Canada), a Senior Researcher at the "Millennium Institute on Foundations of Data" (IMFD, Chile), and a PI at the "National Basal Center for AI Research" (Cenia, Chile). He has Ph.D. in Mathematics from the "Pontificia Universidad Catolica de Chile" (PUC). His broad research interests are related to Data Science and Artificial Intelligence, in particular, explainable AI, knowledge representation, data management, computational logic, ontologies, uncertain reasoning, causality, and machine learning.

Biography

I am a Full Professor in Computer Science at Lyon 1 University (France), affiliated with the Liris research lab. I am the Head of the Database group in the same lab. I am also an Adjunct Professor at the University of Waterloo (Canada) in the Data Systems Group (since 2020). I am an IEEE Senior Member (elevated in 2022).
I am the Program Chair of ACM SIGMOD 2022, a flagship conference on Data Management. I am the Chair of the EDBT Executive Board and Association (2020-2025).
Research Interests: I am interested in the active and interdisciplinary areas of Artificial Intelligence (knowledge graphs, graph databases, and data integration) and their impact on Data-intensive Systems. Precisely, my current research interests are in the interplay of structured (relational) and unstructured (graph-oriented) data paradigms. In particular, I am currently working on data integration and data curation, query learning and query processing, query log analysis, and privacy and security issues for these data models. For some of these topics, I am interested in problems that lead to multi-disciplinary applications (such as those for medical data).

Biography

Mirella Lapata FRSE is a computer scientist and Professor in the School of Informatics at the University of Edinburgh. Working on the general problem of extracting semantic information from large bodies of text, Lapata develops computer algorithms and models in the field of natural language processing (NLP).

Biography

I am the head scientist at Meta AR/VR Assistant, leading the ML efforts in building a smart personal assistant. We innovate and productionize techniques on contextual AI, search, question answering, recommendation, knowledge collection and mining, and multi-modal information management. Prior to joining Meta, I spent nearly a decade working on knowledge graphs at Amazon and Google. Before that, I spent another decade working on data integration and cleaning at AT&T Labs and at the Univ. of Washington, where I received my Ph.D. in Computer Science. I have the great honour to be awarded ACM Distinguished Member for contributions to "data and knowledge integration" and the VLDB Early Career Research Contribution Award for "advancing the state of the art of knowledge fusion".

Biography

Manuela Maria Veloso is the Head of J.P. Morgan AI Research & Herbert A. Simon University Professor in the School of Computer Science at Carnegie Mellon University, where she was previously Head of the Machine Learning Department. She served as president of the Association for the Advancement of Artificial Intelligence (AAAI) until 2014, and was the co-founder and a Past President of the RoboCup Federation. She is a fellow of AAAI, Institute of Electrical and Electronics Engineers (IEEE), American Association for the Advancement of Science (AAAS), and Association for Computing Machinery (ACM). She is an international expert in artificial intelligence and robotics.