Decoupling Exploration and Exploitation for Unsupervised Pre-training with Successor Features

arXiv:2405.02569v1 Announce Type: new
Abstract: Unsupervised pre-training has been on the lookout for the virtue of a value function representation referred to as successor features (SFs), which decouples the dynamics of the environment from the rewards. It has a significant impact on the process of task-specific fine-tuning due to the decomposition. However, existing approaches struggle with local optima due to the unified intrinsic reward of exploration and exploitation without considering the linear regression problem and the discriminator supporting a small …

abstract arxiv cs.ai cs.lg dynamics environment exploitation exploration features fine-tuning function impact pre-training process representation the environment training type unsupervised value