The field of bandit learning and optimization is moving towards addressing complex real-world problems by incorporating fairness, budget constraints, and multi-agent decision-making. Recent developments have focused on designing algorithms that balance exploration and exploitation in various settings, including stochastic multi-armed bandits, restless multi-armed bandits, and distributed multi-agent bandits. Noteworthy papers in this area include 'Revisiting Social Welfare in Bandits: UCB is (Nearly) All You Need', which demonstrates that a standard Upper Confidence Bound (UCB) algorithm can achieve near-optimal Nash regret, and 'Neural Index Policies for Restless Multi-Action Bandits with Heterogeneous Budgets', which introduces a Neural Index Policy for multi-action restless multi-armed bandits with heterogeneous budget constraints. These advancements have the potential to impact various applications, such as clinical trials, energy communities, and online advertising.
Advances in Bandit Learning and Optimization
Sources
Revisiting Social Welfare in Bandits: UCB is (Nearly) All You Need
Cost-Sensitive Freeze-thaw Bayesian Optimization for Efficient Hyperparameter Tuning
Online Mixture of Experts: No-Regret Learning for Optimal Collective Decision-Making
Neural Index Policies for Restless Multi-Action Bandits with Heterogeneous Budgets
Fair Cost Allocation in Energy Communities: A DLMP-based Bilevel Optimization with a Shapley Value Approach
UCB-type Algorithm for Budget-Constrained Expert Learning
Distributed Multi-Agent Bandits Over Erd\H{o}s-R\'enyi Random Networks
Last Iterate Analyses of FTRL in Stochasitc Bandits
Multi-Task Surrogate-Assisted Search with Bayesian Competitive Knowledge Transfer for Expensive Optimization
Infrequent Exploration in Linear Bandits
Engineering Social Optimality via Utility Shaping in Non-Cooperative Games under Incomplete Information and Imperfect Monitoring
Empirical Bayesian Multi-Bandit Learning
Budgeted Multiple-Expert Deferral