Adversarial bandits with noisy side observations

02.05.2022, 10:15  –  2.09.2.22 (hydrid - Zoom)
Forschungsseminar Statistik

Tomas Kocak

In this talk, we take a look at a partial-observability model for online learning problems where the learner, besides its loss, also observes some noisy feedback about the other actions, depending on the underlying structure of the problem. We represent this structure by a weighted directed graph, where the edge weights are related to the quality of the feedback shared by the connected nodes. In the setting without noise, independence number of the graph characterizes the complexity of the problem. In this talk, we define "effective independence number" and show that this quantity characterizes the complexity of the problem.

Please contact us for the zoom link.