Learning in intelligent systems
Hiroshi Makino, Ph.D.
Professor, Department of Physiology
Keio University School of Medicine, Tokyo, Japan
Recent years have seen a resurgence of interplay between artificial intelligence (AI) and neuroscience. While AI offers new theories on how the brain solves complex problems, neuroscience contributes novel algorithms and neural network architectures that can endow machines with cognitive abilities. However, direct comparisons between artificial and biological intelligent systems remain limited. We addressed this gap by examining behaviors and neural representations across multiple domains of intelligence. By training mice and deep reinforcement learning (RL) agents on the same tasks and analyzing the resulting task representations in their respective neural networks, we found that learning in the mouse cortex exhibits key features reminiscent of deep RL algorithms. Furthermore, by deriving theoretical predictions from AI models and empirically testing them in mice, we discovered that the brain composes novel behaviors through a simple arithmetic combination of pre-acquired action-value representations and a stochastic policy. These findings underscore the remarkable parallels in behaviors and neural representations between the two systems and highlight the value of comparative approaches.
References:
