4 March 2021, Computational Social Choice Seminar, Adrian Haret

Abstract

The talk will survey a handful of models dealing with learning in social networks. The setup, in its bare bones, consists of agents arranged in a social network, each agent holding a belief about a ground-truth state while at the same time being subject to influence from other agents it pays attention to, in a pattern captured by the social network. The agents start off from a set of initial beliefs, assumed to be noisy signals about the ground truth, and iteratively update their beliefs by taking into account their neighbors, i.e., agents refine their beliefs by 'learning' from each other. Questions that are of interest with respect to such a process are: whether agents eventually converge to the same collective belief; if this is the case, whether the collective belief accurately tracks the ground truth (a phenomenon called the 'wisdom of the crowds'); and what effect the topology of the network has on issues of convergence and accuracy. There is a vast literature on this topic spanning Economics, Philosophy and Computer Science, but the focus of the talk will be on a selection of prominent models that assume simple and intuitive update rules, i.e., so-called "naïve learning models." Though representing highly stylized versions of real-world interactions, we will see that such models still give rise to interesting learning dynamics.

This talk will be held online. Everybody is welcome. Please note that we will be using the waiting room functionality, so you may have to wait a couple of minutes before getting admitted.

To join, use this link: http://bit.ly/comsoc-illc-21-03-04

For more information on the Computational Social Choice Seminar, please consult https://staff.science.uva.nl/u.endriss/seminar/