From Generalized Linear Bandit to Logistic Bandit: An Overview

Event

Weekly OSI Lab Seminar

Short summary

In this seminar, I will talk about the generalized linear bandits and logistic bandits. Especially for the logistic bandits, I will talk about its connection to the generalized linear bandits and some recent progresses on it from Criteo AI Lab.

Abstract

This talk is divided into two parts. First part deals with the so-called generalized linear bandit, which is the generalized version of the usual linear bandit. The analysis of linear bandit is suitably extended to take into account the nonlinearity caused by arbitrary (inverse) link function \mu by focusing on the reward space rather than the parameter space. Second part deals with logistic bandits, the class of bandits in which a sigmoidal-type nonlinearity is considered, and the output follows the corresponding Bernoulli distribution. This is especially well-fitted with real-life applications in situations where the target is binary (ex. user’s click of an advertisement). Due to high nonlinearity, existing algorithm had a clear trade-off between optimal regret and computational efficiency. Also, as the outputs are binary, naively applying methodologies from the first part leads to suboptimal guarantee. We consider a series of three papers from Criteo AI Lab that sequentially deal with this issue.

Papers

Papers discussed in the seminar:

Part 1

  • Sarah Filippi, Olivier Cappé, Aurélien Garivier, and Csaba Szepesvári. Parametric Bandits: The Generalized Linear Case. In NIPS 2010.

Part 2

  • Louis Faury, Marc Abeille, Clément Calauzènes, and Olivier Fercoq. Improved Optimistic Algorithms for Logistic Bandits. In ICML 2020.
  • Marc Abeille, Louis Faury, and Clément Calauzènes. Instance-Wise Minimax-Optimal Algorithms for Logistic Bandits. In AISTATS 2021.
  • Louis Faury, Marc Abeille, Kwang-Sung Jun, and Clément Calauzènes. Jointly Efficient and Optimal Algorithms for Logistic Bandits. In AISTATS 2022.
Previous
Next