Seminar: Advancing Machine Intelligence Through Reinforcement Learning

Abstract

    A significant portion of learning in animals, including in humans, happens through interactions with their environment. Through interactions, animals are able to learn the consequences of their actions and how to act in order to influence their environment to achieve their goals. Reinforcement learning (RL) provides a computational framework for such goal-directed learning from interactions. The generality of the RL framework allows for the components of a wide range of tasks that are useful in our society across different sectors to be expressed as RL problems. These sectors include healthcare, transportation, energy, business management, and robotics, to name a few. My research aims to address some of the key challenges in building effective solutions for such RL problems. In particular, my research focuses on 1) learning in the absence of informative reward/feedback signals, 2) learning in non-stationary environments where the dynamics of the environment can change over time due to unperceivable causes (including those caused by changes in the behavior of other agents in the environment), and 3) learning in domains with natural language interfaces, where RL agents must learn to understand natural language and also use it to interact with their environment. In this talk, I will first provide an overview of these challenges and then present some of my work on how to address them. I will conclude the talk with a discussion of my planned future work.

Bio

Janarthanan Rajendran is an IVADO postdoctoral research fellow at Mila - Quebec Artificial Intelligence Institute and the University of Montreal, working with Prof. Sarath Chandar and Prof. Doina Precup. He completed his Ph.D. in Computer Science and Engineering at the University of Michigan, Ann Arbor, under the supervision of Prof. Satinder Singh. During his Ph.D., he interned at IBM, Google, and DeepMind. Janarthanan’s research interests lie in building systems that–through interaction–can learn to be competent in complex, dynamic, and uncertain environments. He is interested in computational methods that build such systems as well as their practical applications and societal implications. To this end, his current research mainly focuses on deep reinforcement learning and natural language processing. Janarthanan’s research has been published at top venues such as ICML, NeurIPS, AAAI, ICLR, NAACL, EMNLP, and TACL.