Locality Sensitive Sparse Encoding for Learning World Models Online

arXiv:2401.13034v3 Announce Type: replace
Abstract: Acquiring an accurate world model online for model-based reinforcement learning (MBRL) is challenging due to data nonstationarity, which typically causes catastrophic forgetting for neural networks (NNs). From the online learning perspective, a Follow-The-Leader (FTL) world model is desirable, which optimally fits all previous experiences at each round. Unfortunately, NN-based models need re-training on all accumulated data at every interaction step to achieve FTL, which is computationally expensive for lifelong agents. In this paper, we revisit …

abstract arxiv catastrophic forgetting cs.ai cs.lg data encoding leader networks neural networks nns online learning perspective reinforcement reinforcement learning type world world model world models