Psychophysical detection and learning in freely behaving rats: a probabilistic dynamical model for operant...

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Behavioral Psychology

Background:

• Operant conditioning involves learning associations between stimuli, responses, and outcomes.
• Existing models like Hebbian and Rescorla-Wagner theories offer partial explanations for learning mechanisms.
• Understanding the stochasticity in learning is crucial for accurately modeling animal behavior.

Purpose of the Study:

• To introduce a novel stochastic learning model combining Hebbian and Rescorla-Wagner elements.
• To predict behavioral data from rats performing a vibrotactile yes/no detection task.
• To compare the proposed model's performance against existing reinforcement learning and drift diffusion models.

Main Methods:

• Developed a stochastic learning model with trial-by-trial variability in associative strengths.
• Implemented subject-dependent variance coefficient (SVC) and trial-dependent variance coefficient (TVC) subtypes.
• Utilized information criteria (AIC, BIC) and psychophysical measures (A', hits, false alarms) for model comparison.

Main Results:

• The proposed SVC and TVC models demonstrated superior trial-by-trial fits to experimental data compared to other models.
• The TVC model more closely replicated experimental psychophysical measures, including sensitivity and response biases.
• The model successfully linked psychophysical response measures with learning dynamics in a yes/no detection task.

Conclusions:

• The novel stochastic learning model provides a more accurate account of operant conditioning than previous models.
• The TVC subtype is particularly effective in capturing the probabilistic nature of learning and behavioral variability.
• This approach offers potential applications in neural engineering for understanding and predicting decision-making processes.