Please note: This PhD seminar will be given online.
David
Radke,
PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Supervisors: Professors Kate Larson, Tim Brecht
While it has long been recognized that a team of individual learning agents can be greater than the sum of its parts, recent work has shown that larger teams are not necessarily more effective than smaller ones.
In this talk, I will discuss why and under which conditions certain team structures promote effective learning for a population of individual learning agents. I will show that, depending on the environment, some team structures help agents learn to specialize into specific roles, resulting in more favorable global results. However, large teams create credit assignment challenges that reduce coordination, leading to large teams performing poorly compared to smaller ones. As a result, I will present a preliminary design of an agent that can self-regulate their group alignment to recover favorable learning conditions from a known sub-optimal initialization. To that end, I support the conclusions made in this talk with both theoretical analysis and empirical results.
To join this PhD seminar on Zoom, please go to https://uwaterloo.zoom.us/j/99255853871.