Speaker: Anil Yaman, Vrije Universiteit Amsterdam

Title: Improving the efficiency of collective learning with meta-control of social learning strategies

Synopsis: Natural systems and processes have been one of the main sources of inspiration in the development of artificial intelligence. In particular, the concept of emergence of a global intelligent behavior from the interactions of simple units has been one of the main inspiration in my research. In this talk, I will give some examples where I have explored this idea in simulations of collective systems. Specifically, I will discuss the work where we studied individual and two social learning strategies, namely, success-based and conformist, to provide optimum learning as a collective. First, we showed that environmental uncertainty is a crucial predictor of the performance of these strategies. Then, we proposed a meta-control approach in social learning context to switch between these strategies based on the environmental uncertainty perceived by the individuals. The meta-control approach is motivated by the recent findings in neuroscience showing that the brain can arbitrate between different learning strategies. Our findings show that the meta-control approach can have a significant impact in learning multi-agent systems for resolving environmental uncertainty with minimal exploration cost. 

Bio: Anil Yaman is an Assistant Professor in the Computational Intelligence Group at the Vrije Universiteit Amsterdam. He received his Ph.D. degree in Computer Science from the Eindhoven University of Technology in 2019. Prior to his current position, he held a postdoctoral researcher position at the Center for Neuroscience-inspired AI at Korea Advanced Institute of Science and Technology (KAIST). His main areas of research are evolutionary computation, artificial neural networks, and swarm intelligence. In particular, he is interested in biologically inspired computing models and approaches to learning.

WED 13 APR 2022 4PM – 5PM AEST; 8AM-9AM CEST AMSTERDAM

ZOOM Meeting ID: 873 1557 5255; Password: 778635