Forget-me-some: General versus special purpose models in a hierarchical probabilistic task.

Area of Science:

• Cognitive science
• Computational neuroscience
• Machine learning

Background:

• Human behavior often follows ideal Bayesian observer models in simple environments.
• Complex probabilistic tasks necessitate approximate inference methods due to computational demands.

Purpose of the Study:

• To investigate two approximation schemes for hierarchical inference in a serial reaction time task.
• To compare a general variational method with a novel, task-specific tailored model.

Main Methods:

• Subjects performed a serial reaction time task with stimuli from a hierarchical Markov chain.
• Two computational models were fitted to behavioral data: a general variational Bayesian method and a tailored approximation scheme.

Main Results:

• The tailored approximation model provided a significantly better fit to the behavioral data than the general variational method.
• The tailored model offered new insights into cholinergic manipulation effects.
• Neither model fully explained behavior on more complex contingencies.

Conclusions:

• Human learning in complex tasks appears sensitive to specific task constraints, favoring tailored inference strategies.
• General-purpose inference models have limitations, while task-specific approaches offer valuable perspectives.
• Further research is needed to understand the brain's learning mechanisms and integrate general and specific modeling approaches.