Neural representational geometries reflect behavioral differences in monkeys and recurrent neural networks.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Animals may use diverse strategies for laboratory tasks.
• Combined analysis of behavioral and neural data can obscure individual strategy differences.
• Inferring single-subject strategies is crucial for understanding task performance.

Purpose of the Study:

• To develop and apply techniques for inferring animal strategies at the single-subject level.
• To investigate neural and behavioral differences in monkeys performing a rule-based task.
• To explore the relationship between neural representations, behavior, and training duration.

Main Methods:

• Analysis of behavioral data (performance, reaction times) from two monkeys performing a visually cued rule-based task.
• Examination of neural activity recordings from the dorsolateral prefrontal cortex to analyze stimulus representation geometry.
• Utilizing recurrent neural network models to correlate strategies with training amount.

Main Results:

• Initial behavioral analysis did not reveal strategy differences between monkeys.
• Analysis of neural activity geometry in the dorsolateral prefrontal cortex showed significant differences between the two monkeys.
• Re-analysis of behavior, prompted by neural findings, uncovered differences in reaction times linked to representational geometry.
• These differences suggest the monkeys employed distinct strategies.
• Recurrent neural network models indicated that strategies correlate with training duration.

Conclusions:

• Neural activity patterns can reveal subtle behavioral differences and distinct cognitive strategies not apparent through standard performance metrics.
• The geometry of neural representations provides a powerful tool for inferring individual strategies in animal tasks.
• Training duration may play a significant role in shaping both neural representations and behavioral strategies.