4th workshop on Graphs and more Complex structures for Learning and Reasoning

Human Allied Learning of Neurosymbolic Models

Abstract: Historically, Artificial Intelligence has taken a symbolic route for representing and reasoning about objects at a higher-level or a statistical route for learning complex models from large data. To achieve true AI, it is necessary to make these different paths meet and enable seamless human interaction. First, I will introduce for learning from rich, structured, complex and noisy data. One of the key attractive properties of the learned models is that they use a rich representation for modeling the domain that potentially allows for seam-less human interaction. Next, I will present the recent progress that allows for more reasonable human interaction where the human input is taken as "advice" and the learning algorithm combines this advice with data. Finally, I will discuss about the potential of "closing the loop" where an agent figures out what it knows and solicits information about what it does not know. This is an important direction to realize the true goal of human allied AI.

Speaker Bio: Prof. Sriraam Natarajan is a Professor at the Department of Computer Science at University of Texas Dallas, a hessian.AI fellow at TU Darmstadt and a RBDSCAI Distinguished Faculty Fellow at IIT Madras. His research interests lie in the field of Artificial Intelligence, with emphasis on Machine Learning, Statistical Relational Learning and AI, Reinforcement Learning, Graphical Models and Biomedical Applications. He is a AAAI senior member and has received the Young Investigator award from US Army Research Office, Amazon Faculty Research Award, Intel Faculty Award, XEROX Faculty Award, Verisk Faculty Award and the IU trustees Teaching Award from Indiana University. He is the AAAI program co-chair of AAAI 2024, AI and society track chair of AAAI 2022 and 2023, demo chair of IJCAI 2022, program co-chair of SDM 2020 and ACM CoDS-COMAD 2020 conferences. He was the speciality chief editor of Frontiers in ML and AI journal, and is an associate editor of MLJ, JAIR, DAMI and Big Data journals.

Graph Neural Networks for Financial Data

Abstract: 

Speaker Bio: Mahashweta Das is a Senior Director, Artificial Intelligence at Visa Research where she works on challenging real problems at the crossroads of tech and payment industry. Previously, she was employed as a Research Scientist at Hewlett Packard Labs where she designed and developed big data analytics solutions for HPE’s 'The Machine'. She has also held summer positions at Yahoo! Research, Technicolor Research, and IBM Research. Mahashweta received her Ph.D. in Computer Science from the University of Texas at Arlington in 2013. Her research interests include deep learning, machine learning, data mining, graphs, and algorithms. She has filed over fifteen patents and published over twenty refereed articles at premier international research conferences, and regularly serves on the program committee of these conferences. Her PhD dissertation received Honorable Mention at ACM SIGKDD 2014 Doctoral Dissertation Award. At Visa Research, she is leading a group of AI researchers and engineers focused on conducting foundational research, creating new products/early prototypes that incorporate research breakthroughs, and delivering innovative technologies to Visa's strategic products and businesses. She is also managing the academic collaboration program for Visa Research.

Network geometry detection: why triangles are not enough

Abstract: Geometric network models have been extremely popular, as they capture the natural idea that similar vertices are likely to connect.  Indeed, many real-world networks can be accurately modeled by positioning vertices of a network graph in hyperbolic spaces. However, when observing only the network connections, it may not be known whether the network was generated from an underlying geometry. While triangle counts and clustering coefficients are the standard statistics to signal the presence of geometry, we show that they fail to detect geometry induced by hyperbolic spaces. We, therefore, introduce a different statistic, weighted triangles, which weighs triangles based on their evidence for geometry. We show analytically, as well as on synthetic and real-world data, that weighted triangles are a powerful statistic to detect hyperbolic network geometry.

Speaker Bio: Prof. Clara Stegehuis is an associate professor at Twente Universty. She works at the intersection of probability theory, graph theory and stochastic networks, with an emphasis on asymptotic analysis, stochastic process limits, and randomized algorithms. Problems she investigate are inspired by applications in network science, physics and computer science. Her research interests are Epidemic spreading and percolation, Network motifs, Asymptotic network properties and Geometric networks

Foundation Models for Knowledge Graph Reasoning

Abstract: Foundation models in graph learning are hard to design due to the lack of common invariances that transfer across different structures and domains. In this talk, I will give an overview of ULTRA, our new approach for creating foundation models for knowledge graph reasoning that captures relation interactions and does not require any input node or edge features. Experimentally, a single pre-trained ULTRA in the zero-shot inference mode outperforms supervised SOTA models on 50+ diverse graphs and can generalize to any multi-relational graph. Finally, I will talk about the most recent results in applying ULTRA as a single model for complex logical query answering on KGs.

Speaker Bio: Michael Galkin is an AI Research Scientist at Intel Labs working on Graph ML and Geometric DL. He obtained a double-degree PhD with magna cum laude at the University of Bonn (Germany) and ITMO University (Russia) in 2018, studying knowledge graphs, their data integration strategies and query optimization routines.

AI methods for human networks and high-stakes decisions

Abstract: In an interconnected world, effective policymaking increasingly relies on understanding complex human networks and processes, such as diseases spreading over contact networks, polarization over social networks, and disruptions in supply chains. However, there are many challenges to understanding networks and how they impact decisions, including (1) how to infer human networks from data, when they are only partially observed, (2) how to model complex processes over networks and inform decision-making, (3) how to estimate the causal impacts of decisions, in turn, on human networks. In this talk, I'll discuss how we've addressed each of these challenges in the context of COVID-19 pandemic response, where we've developed new methods for network inference and epidemiological modeling, and deployed decision-support tools for public health officials. I'll also briefly describe our recent work on developing GNNs to model disruptions over supply chains and other dynamic networks.

Speaker Bio: Serina Chang is a final-year PhD student in Computer Science at Stanford University. Her research develops machine learning and network science methods to tackle complex societal challenges, from pandemics to polarization to supply chains. Her work has been published in venues including Nature, PNAS, KDD, AAAI, EMNLP, and ICWSM, and featured by over 650 news outlets, including The New York Times and The Washington Post. Her work is also recognized by the KDD 2021 Best Paper Award, NSF Graduate Research Fellowship, Meta PhD Fellowship, EECS Rising Stars, Rising Stars in Data Science, and Cornell Future Faculty Symposium.

Learning on Graphs: What is Next?

Abstract: Graphs provide a universal representation of data with numerous types. In the last decades, we have witnessed techniques to extract knowledge from graphs for various real-world applications. In this talk, I will first talk about how these techniques evolved, what principles they shared and what unique advances they introduced. Now we are in the era of foundation models. Then I will discuss what progress we have made and what next steps we will potentially take towards graph foundation models.

Speaker Bio: Jiliang Tang is a University Foundation Professor in the computer science and engineering department at Michigan State University. He got one early promotion to associate professor at 2021 and then a promotion to full professor (designated as MSU foundation professor) at 2022. Before that, he was a research scientist in Yahoo Research. He got his Ph.D. from Arizona State University in 2015 and MS and BE from Beijing Institute of Technology in 2010 and 2008, respectively. His research interests include graph machine learning, trustworthy AI, and their applications in Education and Biology. He authored the first comprehensive book “deep learning on graphs” with Cambridge University Press and developed various well-received open-sourced tools including scikit-feature for feature selection, DeepRobust for trustworthy AI and DANCE for single-cell analysis. He was the recipient of various career awards (2022 AI’s 10 to Watch, 2022 IAPR J. K. AGGARWAL, 2022 SIAM SDM, 2021 IEEE ICDM, 2021 IEEE Big Data Security, 2020 ACM SIGKDD, 2019 NSF), numerous industrial faculty awards (Amazon, Cisco, Johnson&Johnson, JP Morgan, Criteo Labs and SNAP), and 8 best paper awards (or runner-ups). He serves as conference organizers (e.g., KDD, SIGIR, WSDM and SDM) and journal editors (e.g., TKDD, TKDE and TOIS). He has published his research in highly ranked journals and top conference proceedings, which have 33,000 citations with h-index 89 and extensive media coverage. More information can be found at https://www.cse.msu.edu/~tangjili/

Practical Challenges in Graph Representation Learning

Speaker Bio: Karthik Subbian is a director at Amazon with more than 20 years of industry experience. He leads a team of scientists and engineers to improve search quality and trust. He was a research leader at Facebook, before coming to Amazon, where he had led a team of scientists and engineers to explore information propagation and user modeling problems using the social network structure and its interactions. Earlier to that, he was working at IBM T.J. Watson research center in the Business Analytics and Mathematical Sciences division. His areas of expertise include machine learning, information retrieval, and large-scale network analysis. More specifically, semi-supervised and supervised learning in networks, personalization and recommendation, information diffusion, and representation learning. He holds a masters degree from the Indian Institute of Science (IISc) and a Ph.D. from the University of Minnesota, both in computer science. Karthik has won numerous prestigious awards, including the IBM Ph.D. fellowship, best paper award at Siam Data Mining (SDM) conference 2013 and Informs Edelman laureate award 2013.