2021/10/11 by Trevor McInroe, Lukas Schäfer, McInroe, Trevor +3 · 2 citations
Computer Science · Mathematics · #Adversarial Robustness in Machine Learning #Artificial Intelligence (cs.AI) #Artificial intelligence #Benchmark (surveying) #Bottleneck #Computer science #Domain Adaptation and Few-Shot Learning #Encoder #FOS: Computer and information sciences #Invariant (physics) #Machine Learning (cs.LG) #Machine learning #Mathematics #Reinforcement Learning in Robotics #Reinforcement learning #Representation (politics) #State space #Task (project management) #cs.AI #cs.LG
paper · pdf · doi:10.48550/arxiv.2110.04935
published in arXiv (Cornell University) (Cornell University)
arxiv created 2021/10/11 · openalex publication_date 2021/10/11 · arxiv updated 2021/10/12 · openalex created_date 2021/10/25 · openalex updated_date 2026/07/28
Deep reinforcement learning (RL) agents that exist in high-dimensional state spaces, such as those composed of images, have interconnected learning burdens. Agents must learn an action-selection policy that completes their given task, which requires them to learn a representation of the state space that discerns between useful and useless information. The reward function is the only supervised feedback that RL agents receive, which causes a representation learning bottleneck that can manifest in poor sample efficiency. We present k-Step Latent (KSL), a new representation learning method that enforces temporal consistency of representations via a self-supervised auxiliary task wherein agents learn to recurrently predict action-conditioned representations of the state space. The state encoder learned by KSL produces low-dimensional representations that make optimization of the RL task more sample efficient. Altogether, KSL produces state-of-the-art results in both data efficiency and asymptotic performance in the popular PlaNet benchmark suite. Our analyses show that KSL produces encoders that generalize better to new tasks unseen during training, and its representations are more strongly tied to reward, are more invariant to perturbations in the state space, and move more smoothly through the temporal axis of the RL problem than other methods such as DrQ, RAD, CURL, and SAC-AE.