What can non-linear embeddings tell us about the way a mouse learns a motor skill?

ALVARO CONCHA; JORGE MIRANDE; LEONARDO MOLANO RAMIREZ; DAMIAN HERNANDEZ; MARIA SOLEDAD ESPOSITO

Virtual

Congreso; Sociedad Argentina de Investigación en Neurociencias; 2021

Animals exhibit complex behavioral repertoires that can be described as combinations from a finite set of stereotyped movements. These be- haviors are flexible, since different movement sequences can be used to solve similar tasks, and are adaptable to changing environments through learning mechanisms.We used unsupervised machine learning to classify different types of movements executed by mice performing a motor skill learning task. We constructed UMAP embeddings to find a low dimensional representation of mouse be- havior. Then, we clustered behaviors into sepa- rate categories and studied their changes with training and between subjects.