Infinite hidden Markov models can dissect the complexities of learning.

Area of Science:

• Computational Neuroscience
• Behavioral Science
• Machine Learning

Background:

• Task learning is complex and varies significantly between individuals.
• Characterizing learning curves requires models that capture both emergent behaviors and subtle adaptations.
• Existing models may not fully capture the dynamic nature of behavioral learning.

Purpose of the Study:

• To develop a novel computational model for quantitatively characterizing animal learning.
• To capture both the emergence of new behaviors and the adaptation of existing ones during task learning.
• To provide a tool for comprehensive behavioral analysis during learning processes.

Main Methods:

• Development of a dynamic infinite hidden semi-Markov model (DIHSM).
• The DIHSM uses latent states to represent specific behavioral components.
• Model fitting to behavioral data from over 100 mice learning a contrast-detection task.

Main Results:

• The DIHSM successfully described new behaviors by introducing new states and adaptations through state dynamics.
• Analysis of mouse data revealed large interindividual differences in learning.
• Most mice exhibited three distinct stages of task understanding during learning.

Conclusions:

• The proposed dynamic infinite hidden semi-Markov model is effective for capturing complex learning dynamics.
• Behavioral learning in mice progresses through identifiable stages, with new behaviors often appearing at session onset.
• The model offers a powerful new tool for comprehensive analysis of behavioral data during learning.