Regulation of reinforcement learning parameters captures long-term changes in rat behaviour.

Area of Science:

• Neuroscience
• Computational Biology
• Animal Behavior

Background:

• Animals in unpredictable environments face a learning dilemma: exploit current knowledge or explore for new information.
• The regulation of this explore-exploit trade-off during initial learning remains poorly understood.

Purpose of the Study:

• To investigate how rats adapt their reinforcement learning strategies over time during task acquisition.
• To compare computational models for explaining long-term behavioral changes in learning.

Main Methods:

• Observed 24 rats over 24 days on a three-armed bandit task.
• Analyzed daily changes in rat performance and win-shift tendency.
• Utilized computational modeling, including meta-learning approaches.

Main Results:

• Rat performance and win-shift tendencies showed progressive stabilization across pretraining days.
• Behavioral adaptations were successfully modeled by a meta-learning approach.
• The model indicated that learning rate or inverse temperature was regulated by average reward rate.

Conclusions:

• Rats progressively tune their reinforcement learning parameters during initial task learning.
• Meta-learning provides a framework for understanding adaptive learning in uncertain environments.
• Average reward rate is a key factor in regulating exploration-exploitation dynamics.