Information Theory & RL Reading List

Curated by Mouhssine Rifaki | Stanford Electrical Engineering | Last updated August 2026

Information theory provides a principled language for regularization, exploration, and representation in reinforcement learning. This list traces the arc from Shannon and the information bottleneck to empowerment, variational intrinsic control, and modern maximum-entropy methods.

Information Theory & RL: 10 key papers

1. Deep Learning and the Information Bottleneck Principle
   Tishby and Zaslavsky. arXiv 2015.
2. Deep Variational Information Bottleneck
   Alemi et al. arXiv 2016.
3. The information bottleneck method
   Tishby et al. arXiv 2000.
4. Information-Theoretic Confidence Bounds for Reinforcement Learning
   Lu and Van Roy. arXiv 2019.
5. InfoBot: Transfer and Exploration via the Information Bottleneck
   Goyal et al. arXiv 2019.
6. MINE: Mutual Information Neural Estimation
   Belghazi et al. arXiv 2018.
7. On Variational Bounds of Mutual Information
   Poole et al. arXiv 2019.
8. Learning to Optimize via Information-Directed Sampling
   Russo and Van Roy. arXiv 2014.
9. An Information-Theoretic Analysis of Thompson Sampling
   Russo and Van Roy. arXiv 2014.
10. Dynamic Bottleneck for Robust Self-Supervised Exploration
    Bai et al. arXiv 2021.

Related reading lists

• Intrinsic Motivation & Empowerment
• Representation Learning for RL
• Multi-Armed Bandits & Best-Arm Identification
• Bayesian RL & Thompson Sampling
• Variational Inference & ELBO
• Communication in Multi-Agent RL
• Zero-Shot RL & Generalization
• Curiosity-Driven Exploration