Timezone: »

 
Spotlight
Decoupling Representation Learning from Reinforcement Learning
Adam Stooke · Kimin Lee · Pieter Abbeel · Michael Laskin

Thu Jul 22 06:40 AM -- 06:45 AM (PDT) @
in Representation Learning 2 »

In an effort to overcome limitations of reward-driven feature learning in deep reinforcement learning (RL) from images, we propose decoupling representation learning from policy learning. To this end, we introduce a new unsupervised learning (UL) task, called Augmented Temporal Contrast (ATC), which trains a convolutional encoder to associate pairs of observations separated by a short time difference, under image augmentations and using a contrastive loss. In online RL experiments, we show that training the encoder exclusively using ATC matches or outperforms end-to-end RL in most environments. Additionally, we benchmark several leading UL algorithms by pre-training encoders on expert demonstrations and using them, with weights frozen, in RL agents; we find that agents using ATC-trained encoders outperform all others. We also train multi-task encoders on data from multiple environments and show generalization to different downstream RL tasks. Finally, we ablate components of ATC, and introduce a new data augmentation to enable replay of (compressed) latent images from pre-trained encoders when RL requires augmentation. Our experiments span visually diverse RL benchmarks in DeepMind Control, DeepMind Lab, and Atari, and our complete code is available at \url{https://github.com/astooke/rlpyt/tree/master/rlpyt/ul}.

Keywords: Online Learning · Submodular Optimization · Algorithms -> Bandit Algorithms; Algorithms · Embedding and Representation learning

Author Information

Adam Stooke (UC Berkeley)
Kimin Lee (UC Berkeley)
Pieter Abbeel (UC Berkeley & Covariant)
Michael Laskin (UC Berkeley)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors