Conquering OpenAI Retro Contest 3: A Collection of Related Papers

OpenAI Retro Contest is a new kind of Reinforcement Learning benchmark targeting on Transfer Learning, Multi-Task Learning and Meta Learning. In this blog, I make a collection of related papers, so that we can dig deeper to find something valuable for the contest.

[1] Learn to Reinforcement Learn [pdf]

[2] RL²: Fast Reinforcement Learning via Slow Reinforcement Learning [pdf]

[3] Actor-Mimic: Deep Multitask and Transfer Reinforcement Learning [pdf]

[4] Distral: Robust Multitask Reinforcement Learning [pdf]

[5] PathNet: Evolution Channels Gradient Descent in Super Neural Networks [pdf]

[6] Progressive Neural Networks [pdf]

[7] Some Considerations on Learning to Explore via Meta-Reinforcement Learning [pdf]

[8] DARLA: Improving Zero-Shot Transfer in Reinforcement Learning [pdf]

[9] Universal Agent For Disentangling Environments And Tasks [pdf]