The event is organised remotely via zoom: https://aalto.zoom.us/j/67006038782

Abstract: Hierarchical reinforcement learning (HRL) has been proposed as an effective and scalable way to manage large learning problems. It does this by abstracting tasks into subtasks with their individual goals and policies. My talk investigates the cognitive scientific plausibility of HRL as a model of supervisory control in the human mind. Humans frequently face the problem of resource allocation, where multiple, potentially conflicting tasks compete for the same cognitive resources. I discuss the neurocomputational basis of supervisory control in human multitasking and show how the paradigm of computational rationality can converge it with HRL. Finally, I show how to apply this technique to model human multitasking in driving.

Bio: Jussi P.P. Jokinen received his Ph.D. in cognitive science in 2015 from University of Jyväskylä, Finland. His work deals with computational cognitive models of adaptive behaviour. He is a postdoc researcher at User Interfaces group, Aalto University, where his models are employed to understand the cognitive aspects of human-computer interaction. His long-term research passion is in creating psychologically plausible simulation models of human cognition and behaviour from the principles of computational rational analysis.

Speaker: Dr. Jussi Jokinen

Affiliation: Department of Communications and Networking, Aalto University

Machine Learning Coffee Seminar is organized weekly by the Finnish Center for Artificial Intelligence FCAI.