On the Computational Tractability of Robust Bandits

Key points

• Learning when the environment does not belong to the learner's hypothesis class is typically handled using agnostic learning guarantees.1
• However, for anything beyond supervised learning, agnostic guarantees are difficult to come by.1
• In this paper we identify a special case that admits a polynomial-time learner with $\tilde{O}(\sqrt{T})$ regret.1
• It has been recently suggested (Kosoy, 2018) that computationally efficient learners for unrealizable learning problems are crucial for solving the AI alignment problem.1

Sources (1)

• [1]On the Computational Tractability of Robust Bandits
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 05:38 PM

  Recently, imprecise bandits (Kosoy, 2025) (later renamed to robust bandits in Appel and Kosoy, 2025) were introduced as another approach to unrealizable learning in the bandits setting and a $Θ(\sqrt{T})$ regret learner was shown for a large class.

  Learning when the environment does not belong to the learner's hypothesis class is typically handled using agnostic learning guarantees.

Extractive summary: sentences quoted from the sources.