Event
Two Reinforcement Learning Problems from Healthcare and Generative Social Science
Virtual Informal Systems Seminar (VISS) Centre for Intelligent Machines (CIM) and Groupe d'Etudes et de Recherche en Analyse des Decisions (GERAD)
Meeting ID: 910 7928 6959 聽 聽 聽 聽
Passcode: VISS
Speaker: Jayakumar Subramanian, Senior Research Scientist, 聽Media and Data Science Research Lab at Adobe (India)
Abstract:
I will talk about two papers in my talk in addition to giving a brief overview of research at the Media and Data Science Research Lab at Adobe India. The first part of my talk covers reinforcement Learning (RL) in sequential estimation and prediction problems in healthcare. In practice, successful RL relies on informative latent states derived from sequential observations to develop optimal treatment strategies. How best to construct such states in a healthcare setting is an open question. In this work, we perform an empirical study of several information encoding architectures using data from septic patients in the MIMIC-III dataset to form representations of a patient state. We find that sequentially formed state representations (including the approximate information state approach) 聽facilitate effective policy learning in batch settings, validating a more thoughtful approach to representation learning that remains faithful to the sequential and partial nature of healthcare data. The second part of my talk covers use of agent based models (ABM) and machine learning frameworks for inverse generative social science problems. Conventional ABM frameworks are inefficient for large populations and do not differentiate between agent transition modeling and agent behavior modeling, which makes behavior learning in these frameworks challenging.聽 To overcome these problems, we have developed the DeepABM framework which takes a network-centric functional architecture and is built using the concepts of graph convolutional neural networks from deep learning frameworks. Using graph convolutional networks has enabled the following key benefits in DeepABM: i) scale ABMs to large agent populations in real-time, ii) run ABMs efficiently on GPUs, and iii) enable more efficient calibration of ABMs using gradient-based supervised machine learning instead of the status-quo randomized search methods.
叠颈辞:听
Jayakumar is a Senior Research Scientist at the Media and Data Science Research Lab at Adobe, India. His research interests include reinforcement learning in single and multi-agent systems. He has a Ph.D. in reinforcement learning in partially observed and multi-agent systems from 苹果淫院 and dual degrees (Bachelor + Master) in Aerospace Engineering from the Indian Institute of Technology, Bombay.