Paper
A Consciousness-Inspired Planning Agent for Model-Based Reinforcement\n Learning
We present an end-to-end, model-based deep reinforcement learning agent which\ndynamically attends to relevant parts of its state during planning. The agent\nuses a bottleneck mechanism over a set-based representation to force the number\nof entities to which the agent attends at each planning step to be small. In\nexperiments, we investigate the bottleneck mechanism with several sets of\ncustomized environments featuring different challenges. We consistently observe\nthat the design allows the planning agents to generalize their learned\ntask-solving abilities in compatible unseen environments by attending to the\nrelevant objects, leading to better out-of-distribution generalization\nperformance.\n
Authors: Zhao, Mingde · Liu, Zhen · Luan, Sitao · Zhang, Shuyuan · Precup, Doina · Bengio, Yoshua