Anonymous Bandits for Multi-User Systems

TL;DR

This paper introduces anonymous bandits, a new online learning framework for multi-user systems that ensures user privacy through k-anonymity, and provides the first algorithms with sublinear regret bounds.

Contribution

It extends bandit algorithms to satisfy k-anonymity constraints, enabling online user clustering without individual user data, and establishes regret bounds for this setting.

Findings

First sublinear regret algorithms for anonymous bandits

Lower bounds demonstrating the difficulty of the problem

Framework enables privacy-preserving online learning

Abstract

In this work, we present and study a new framework for online learning in systems with multiple users that provide user anonymity. Specifically, we extend the notion of bandits to obey the standard $k$-anonymity constraint by requiring each observation to be an aggregation of rewards for at least $k$ users. This provides a simple yet effective framework where one can learn a clustering of users in an online fashion without observing any user's individual decision. We initiate the study of anonymous bandits and provide the first sublinear regret algorithms and lower bounds for this setting.