Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster A in Poster Session A: Tuesday, August 4, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Hidden Markov Modeling of Labyrinth Navigation

Zeyu Jing¹, Jiang Wu¹, Markus Meister¹; ¹California Institute of Technology

Presenter: Zeyu Jing

Naturalistic maze navigation is complex and requires unsupervised models for segmentation of behavior. Here, we model animals’ navigation behavior using a Hidden Markov Model (HMM) that predicts turning actions from spatial locations. The resulting HMM outperforms a vanilla Markov model and generalizes across animals. The inferred hidden states are interpretable, capturing distinct behavioral policies, including homing, reward-seeking, and exploration. Interestingly, the transition structure among these states exhibits strong cyclicity, suggesting periodic sequencing of behavioral policies. To capture learning-related nonstationarity, we developed a hierarchical extension that identifies an abrupt shift from a naive to an expert stage, coinciding with animals' sudden performance improvement. Together, these results suggest that animals learn to navigate complex environments by acquiring a small set of efficient navigation policies and cycle systematically among them.

Topic Area: Methods, Tools, Theory & Neural Coding