Context-dependent decision-making: a simple Bayesian model.

Area of Science:

• Cognitive Science
• Animal Behavior
• Computational Neuroscience

Background:

• Animal learning is often explained by context-learning processes.
• Animals adapt behavior to changing environmental contexts.
• Understanding context inference is key to explaining animal decision-making.

Purpose of the Study:

• To present a novel computational model of sequential context inference in animal learning.
• To approximate full Bayesian inference using a sequential-sampling scheme.
• To investigate how animals adapt behavior to changing environmental contexts.

Main Methods:

• A novel decision-making model using a Chinese restaurant process with inertia.
• Approximation of Bayesian inference via a sequential-sampling scheme.
• Thompson sampling for action selection to drive exploration.

Main Results:

• The model was tested on two-alternative choice tasks with switching reinforcement schedules.
• Model predictions were compared with rat behavioral data from T-maze studies.
• The model successfully replicated spontaneous recovery, partial reinforcement effects, and reversal learning.

Conclusions:

• The proposed model provides a unified framework for understanding context-learning phenomena in animals.
• The model's ability to replicate diverse behavioral effects highlights its explanatory power.
• This work advances computational approaches to animal cognition and decision-making.