Learning values across many orders of magnitude
This addresses a scaling issue in reinforcement learning for AI agents, but it is incremental as it builds on prior work to replace a heuristic.
The paper tackles the problem of learning algorithms being sensitive to the scale of the function being approximated, particularly in value-based reinforcement learning where value magnitudes change over time. It proposes adaptive normalization of targets, enabling removal of reward clipping in Atari games without performance loss.
Most learning algorithms are not invariant to the scale of the function that is being approximated. We propose to adaptively normalize the targets used in learning. This is useful in value-based reinforcement learning, where the magnitude of appropriate value approximations can change over time when we update the policy of behavior. Our main motivation is prior work on learning to play Atari games, where the rewards were all clipped to a predetermined range. This clipping facilitates learning across many different games with a single learning algorithm, but a clipped reward function can result in qualitatively different behavior. Using the adaptive normalization we can remove this domain-specific heuristic without diminishing overall performance.