Recording available to view at: https://mediaspace.illinois.edu/media/t/1_fvpqstp8
Abstract:
Modeling complex systems that represent problems in the social, health, and cognitive sciences require techniques that can identify variable interactions of non-linear and sparse domain. There are challenges in representation, computation (inference and predictions), and interpretation - particularly when the models use latent variables. In my talk, I will discuss a few projects that provide insights about how to solve these three issues. The tools I develop arise from relational learning and probabilistic generative models (PGMs). First, I will describe a problem related to using parametric redundancy and probabilistic mass allocation in a PGM that models sparse data and how this redundancy could be used to facilitate learning computations for a family of probabilistic generative models. Second, I will present an alternative to abstract the latent structure of discrete variables by representing their interactions using relational approaches for high dimensional sparse matrices through hierarchical organization. Third, I will discuss some applications of my methodological work where I investigate the use of time series data models and how their complex interactions can be used to predict individual attributes. Fourth, identifying the laws that govern a complex system based on its network representation involves a lossy compression of the system. I will discuss strategies to represent changes in a dynamical system using its footprint data by taking into account this loss of information. I will illustrate the applications of these techniques to social science, health informatics, and cognitive science.
Bio:
Dr. Pablo Robles-Granda is an Illinois Future Faculty Fellow in Computer Science with a joint appointment in the Psychology Department at the University of Illinois at Urbana-Champaign. He is a member of the Roster of Experts of the World Health Organization Digital Health Technical Advisory Group. His research focuses on probabilistic and relational approaches to machine Learning and statistics and their applications on observational and experimental data in Health Informatics, Cognitive Science, and Neuroscience. Previously he was a Research Assistant Professor in computer science and engineering at the University of Notre Dame. He received his PhD from Purdue University. He was a Top-Reviewer at the International Conference on Machine Learning (ICML 2020), a SIGKDD Student Scholar at the ACM’s Celebration of 50 Years of the A. M. Turing Award, a recipient of the Excellence Fellowship from the National Science Foundation of Ecuador, a recipient of an Academic Merit Recognition from the Office of International Programs and Services and the Hispanic Association at SIUC among other awards. He regularly serves on the board of several organization and program committees of top-tier conferences including AAAI, ICML, NeurIPS, and many others. He serves as reviewer for several of the main journals in machine learning and data science (TPAMI, DAMI, TKDE, KAIS).
Part of the Illinois Computer Science Speakers Series. Faculty Host: Sanmi Koyejo
