Strategy inference during learning via cognitive activity-based credit assignment models.

Area of Science:

• Neuroscience
• Behavioral Science
• Machine Learning

Background:

• Understanding individual learning strategies is crucial for personalized education and cognitive science.
• Current methods often require large datasets or predefined strategy models.
• Behavioral data offers a rich, albeit complex, source for inferring cognitive processes.

Purpose of the Study:

• To develop a computational method for inferring individual learning strategies from behavioral data alone.
• To apply this method to identify specific strategies in a rodent learning task.
• To validate the identified strategy using neurophysiological recordings.

Main Methods:

• Developed a novel computational framework using Activity-Credit Assignment algorithms.
• Integrated a statistical hold-out selection method for strategy identification.
• Applied the method to behavioral data from rats in a continuous T-maze task.
• Collected and analyzed neuronal data from the dorsomedial striatum.

Main Results:

• The method successfully identified a distinct learning strategy based solely on behavioral data.
• The identified strategy involves 'chunking' previously learned paths.
• Neuronal activity in the dorsomedial striatum correlated with this path-chunking strategy.
• Demonstrated the efficacy of the computational approach in a real-world learning scenario.

Conclusions:

• Individual learning strategies can be inferred computationally from behavioral data.
• Path chunking is a relevant learning strategy in spatial navigation tasks for rats.
• The dorsomedial striatum plays a role in implementing this chunking strategy.
• This approach offers a powerful tool for studying learning and cognition at the individual level.