<ul><li>Learning in environments with sparse rewards remains a fundamental challenge in reinforcement learning.</li><li>Artificial curiosity addresses this limitation through intrinsic rewards to guide exploration.</li><li>Leveraging information geometry, intrinsic rewards can be constrained to concave functions of the reciprocal occupancy.</li><li>This framework integrates foundational exploration methods into a single, cohesive model.</li></ul>

An Information-Geometric Approach to Artificial Curiosity

Discover more