Contextual Stochastic Bandits with Budget Constraints and Fairness Application

Explore the fundamental concepts and recent advancements in contextual stochastic bandits with budget constraints and fairness applications in this 47-minute lecture. Delve into the core principles of contextual stochastic bandits, where vector-valued contexts and actions yield stochastic rewards based on partially unknown functions. Learn about the challenge of balancing exploration and exploitation to maximize cumulative rewards and minimize regret. Examine the extension to contextual bandits with knapsacks, incorporating budget constraints and costs. Discover how cost modeling can be applied to fairness constraints in machine learning applications. Gain insights into a novel strategy for handling cost constraints over rounds, specifically designed for fairness applications. Conclude with a discussion of recent research findings from a collaborative study presented at NeurIPS 2023, focusing on small total-cost constraints in contextual bandits with knapsacks and their application to fairness.