Mice exhibit stochastic and efficient action switching during probabilistic decision making.

Area of Science:

• Neuroscience
• Behavioral Economics
• Computational Biology

Background:

• Decision-making in uncertain environments requires integrating internal and external cues.
• Understanding animal decision-making provides insights into fundamental cognitive processes.

Purpose of the Study:

• To model mouse choice behavior in a probabilistic, nonstationary environment.
• To investigate the mechanisms underlying action switching in decision-making tasks.

Main Methods:

• Trained mice on a two-armed bandit task with time-varying reward probabilities.
• Comprehensively modeled choice and action switching behavior using logistic regression, reinforcement learning, and sticky Bayesian inference.

Main Results:

• Mouse behavior exhibited both deterministic and stochastic elements, deviating from optimal Bayesian inference.
• Equivalent models accurately described mouse behavior, incorporating stochastic policies, history-dependent action values, and action repetition.
• Models captured near-maximal reward rates with reduced action switching by adjusting a 'stickiness' parameter.

Conclusions:

• Mouse decision-making in changing environments achieves near-optimal performance with minimal action switching.
• Behavior can be parsimoniously explained by a set of mathematically equivalent models with few parameters.
• These findings offer a framework for understanding flexible decision-making strategies.