Deep Reinforcement Learning Study Group

This repository is a clone of the Spring 2017 deep reinforcement learning class at Berkeley. We've cloned it for the purpose of having a collaborative study group watching the lectures and working on the problem sets.

Resources.

Lecture Material

• Google's deep learning Udacity course seems like a good introduction.

• Professor Bertsekas's Dynamic Programming lectures seem like a good supplement to these lectures.

Papers

• Yann Lecun's paper discusses tricks for normalization and initialization.

Blog Posts etc.

• A post describing RNNs.
• A post explaining LSTMs.
• A pair of posts on seq2seq and attention.
• A post giving general background on RL and diving into score function gradient estimators.

Books

• The Deep Learning Book.

Software

Tensorflow

• Tutorials

• Tensorboard seems like a good way to instrument training algorithms.

• Supervisor

• Threading and queues

• TF Slim might save time.

Other

• Pudb seems like a useful debug tool.

Other courses

• Stanford CS class CS231n: Convolutional Neural Networks for Visual Recognition - another class more focused on convnets