Backprop-Free Reinforcement Learning with Active Neural Generative Coding

In humans, perceptual awareness facilitates the fast recognition and extraction of information from sensory input. This largely depends on how human agent interacts with environment. this work, we propose active neural generative coding, a computational framework for learning action-driven models without backpropagation errors (backprop) in dynamic environments. Specifically, develop an intelligent that operates even sparse rewards, drawing inspiration cognitive theory planning as inference. We demonstrate several simple control problems our performs competitively deep Q-learning. The robust performance offers promising evidence backprop-free approach inference can drive goal-directed behavior.