Leveraging Side Observations in Stochastic Bandits

Abstract

This paper considers stochastic bandits with side observations, a model that accounts for both the exploration/exploitation dilemma and relationships between arms. In this setting, after pulling an arm i, the decision maker also observes the rewards for some other actions related to i. We will see that this model is suited to content recommendation in social networks, where users' reactions may be endorsed or not by their friends. We provide efficient algorithms based on upper confidence bounds (UCBs) to leverage this additional information and derive new bounds improving on standard regret guarantees. We also evaluate these policies in the context of movie recommendation in social networks: experiments on real datasets show substantial learning rate speedups ranging from 2.2x to 14x on dense networks.

Appears in Proceedings of the Twenty-Eighth Conference on Uncertainty in Artificial Intelligence (UAI2012)

References in corpus (4)

• A Contextual-Bandit Approach to Personalized News Article Recommendation
• A Gradient Descent Algorithm on the Grassman Manifold for Matrix Completion
• Bandit Problems with Side Observations
• From Bandits to Experts: On the Value of Side-Observations

Cited by in corpus (10)

• Revealing graph bandits for maximizing local influence
• Reinforcement Learning with Feedback Graphs
• Bounded Regret for Finitely Parameterized Multi-Armed Bandits
• Online learning with Erdős-Rényi side-observation graphs
• Experimental Design for Regret Minimization in Linear Bandits
• When to Reset Your Keys: Optimal Timing of Security Updates via Learning
• Strategic Arms with Side Communication Prevail Over Low-Regret MAB Algorithms
• Covariance Adaptive Best Arm Identification
• Bandit based centralized matching in two-sided markets for peer to peer lending
• Thompson Sampling for Unimodal Bandits