Non-stochastic Bandits With Evolving Observations

TL;DR

This paper introduces a new online learning framework for non-stochastic bandits with evolving, adversarial feedback, providing algorithms with regret bounds that adapt to feedback accuracy and unify several existing models.

Contribution

It presents a unified framework for non-stochastic bandits with evolving observations and develops regret minimization algorithms with novel bounds.

Findings

Algorithms match known regret bounds in special cases

New regret bounds are established for evolving feedback scenarios

Framework generalizes delayed and corrupted feedback models

Abstract

We introduce a novel online learning framework that unifies and generalizes pre-established models, such as delayed and corrupted feedback, to encompass adversarial environments where action feedback evolves over time. In this setting, the observed loss is arbitrary and may not correlate with the true loss incurred, with each round updating previous observations adversarially. We propose regret minimization algorithms for both the full-information and bandit settings, with regret bounds quantified by the average feedback accuracy relative to the true loss. Our algorithms match the known regret bounds across many special cases, while also introducing previously unknown bounds.