Choisir la langue :

Cleaning up the Neighbourhood: A Full Classification for Adversarial Partial Monitoring

Institutional tag: 
Thematic tag(s): 

Partial monitoring is a generalization of the multi-armed bandit framework that decouples the loss from the observations. The setting is sufficiently generic to model full information problems, bandit problems and other settings between and beyond these extremes. I will introduce the setup and describe a new algorithm that greatly simplifies the analysis. Finally I'll describe some of the open problems. This is joint work with Csaba Szepesvari.

 

Dates: 
Friday, March 9, 2018 - 11:00
Location: 
Inria, room A00
Speaker(s): 
Tor Lattimore
Affiliation(s): 
DeepMind