Contextual Information-Directed Sampling
This work addresses a specific design issue in reinforcement learning algorithms for contextual bandits, offering an incremental improvement in data efficiency.
The paper tackled the problem of selecting the appropriate information ratio for information-directed sampling (IDS) in contextual bandit settings, demonstrating that contextual IDS outperforms conditional IDS by considering context distribution, with empirical validation on a neural network contextual bandit.
Information-directed sampling (IDS) has recently demonstrated its potential as a data-efficient reinforcement learning algorithm. However, it is still unclear what is the right form of information ratio to optimize when contextual information is available. We investigate the IDS design through two contextual bandit problems: contextual bandits with graph feedback and sparse linear contextual bandits. We provably demonstrate the advantage of contextual IDS over conditional IDS and emphasize the importance of considering the context distribution. The main message is that an intelligent agent should invest more on the actions that are beneficial for the future unseen contexts while the conditional IDS can be myopic. We further propose a computationally-efficient version of contextual IDS based on Actor-Critic and evaluate it empirically on a neural network contextual bandit.