Reinforcement Learning

Realizations

• Old experiments on RL (2016)
• Solving OpenAI Gym environments (2017-2018)
• Developing an multi agent Tic Tac Toe environment and solving it with Policy Gradients (May 2017)
• Using RL to automatically adapt the cooling in a Data Center (August 2017)
• Controlling Robots via Reinforcement Learning (November 2017)
• Playing and solving the Chrome Dinosaur Game with Evolution Strategies and PyTorch (January 2018)
• Delivery optimization using Reinforcement Learning (January 2019)
• Rubik's Cube optimization (February 2019)
• Multi-Agents simulations (November 2019)

Libraries

• rl is a simple library to do Reinforcement Learning with Keras, it uses old Keras versions and should be updated
• hyperion is a simple multi agent simulation library

References and inspiration

RL references

• Udemy course on RL
• David Silver course on RL at UCL
• Berkeley course on AI
• Spinning up course by OpenAI

Q Learning references

• Q Learning tutorial by Arthur Juliani
• Q Learning tutorial on Keon.io
• Q Learning tutorial by Udacity

Deep Q Learning

• David Silver's Deep Q Learning course
• Demystyfing Deep Reinforcement Learning
• Siraj Raval's notebook on Deep Q Learning

Policy Gradient

• Deep Reinforcement Learning: Pong from Pixels Andrej Karpathy's blog article on RL (always a reference)

Evolution strategies

• Evolution strategies - OpenAI
• How evolution taught us the “genetic algorithm”
• Making a robot learn how to move, part 1 — Evolutionary algorithms
• Optimize a quadratic function with ES - Andrej Karpathy
• Evolution modelling with creatures
• Genetic biwalkers
• Evolving stable strategies

Actor Critic, A2C, ACKTR

• A3C tutorial tutorial by Arthur Juliani
• A3C tutorial with Keras and OpenAI
• A3C explananations and implementations
• ACKTR & A2C - by OpenAI
• ACKTR & A3C implementation in PyTorch
• Actor Critic model with Keras
• Car Racing solving with A3C and this solution as well

PPO, TRPO

• Proximal Policy Optimization - by OpenAI
• PPO,TRPO tutorials

AlphaGo

• ELI5 MCTS
• How AlphaGo works
• Original Paper for AlphaGo by David Silver

Monte Carlo Tree Search

• Udacity videos on MCTS

Misc

• Learning to optimize with RL

Environment

• Unity Agents
• SerpentAI
• Pybullet

Papers

• Discrete Sequential Prediction of Continuous Actions for Deep RL
• Emotion in Reinforcement Learning Agents and Robots: A Survey
• Combating Reinforcement Learning's Sisyphean Curse with Intrinsic Fear
• Curiosity-driven Exploration by Self-supervised Prediction
• End-to-end optimization of goal-driven and visually grounded dialogue systems
• Deep reinforcement learning from human preferences - OpenAI
• Programmable Agents - Deepmind
• Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments - OpenAI
• Actor-Critic Reinforcement Learning with Simultaneous Human Control and Feedback
• Noisy Networks for Exploration
• Hindsight Experience Replay
• DARLA: Improving Zero-Shot Transfer in Reinforcement Learning
• Leveraging Demonstrations for Deep Reinforcement Learning on Robotics Problems with Sparse Rewards
• Evolution Strategies as a Scalable Alternative to Reinforcement Learning
• A Distributional Perspective on Reinforcement Learning
• Intrinsically Motivated Goal Exploration Processes with Automatic Curriculum Learning
• Asynchronous Methods for Deep Reinforcement Learning
• Value Iteration Networks
• A deep reinforcement learning chatbot - MILA
• The Uncertainty Bellman Equation and Exploration
• Deep Reinforcement Learning that Matters
• Overcoming Exploration in Reinforcement Learning with Demonstrations
• Using Simulation and Domain Adaptation to Improve Efficiency of Deep Robotic Grasping
• Rainbow: Combining Improvements in Deep Reinforcement Learning
• Optimizing Long Short-Term Memory Recurrent Neural Networks UsingAnt Colony Optimization to Predict Turbine Engine Vibration
• Continuous Adaptation via Meta-Learning in Nonstationary and Competitive Environments
• Emergent Complexity via Multi-Agent Competition
• A Unified Game-Theoretic Approach to Multiagent Reinforcement Learning