A new arXiv paper analyzes KL-regularized contextual bandits under both reward and preference feedback. The authors show that a greedy sampling approach attains logarithmic regret without an explicit dependence on the eluder dimension. The work covers regret guarantees for the reward-feedback setting and extends the analysis to preference-based feedback.
A new arXiv paper studies stochastic linear contextual bandits where the set of available actions can vary arbitrarily, depending on both the unknown parameter and past interactions. The authors prove matching upper and lower bounds on regret that agree up to logarithmic factors, characterizing the problem's minimax rate.
A new arXiv preprint proposes COBRA-Skills, a method that applies contextual bandit guidance to evolve reusable skills for large language model agents. The approach aims to cut the reliance on expensive execution-based evaluation and large task datasets that limit existing skill optimization techniques. It targets agents that reuse skills distilled from earlier task experience.
A new arXiv paper presents Meta-LinEXP3, an algorithm that brings meta-learning to adversarial linear contextual bandits, a setting where earlier results largely covered only stochastic or non-contextual variants. The method uses an online-within-online framework to transfer knowledge across a sequence of bandit tasks.