Artificial Intelligence-based Study Of Animals' Structured Behavior

The objective quantization of animal behavior is a necessary step in the study adopting animal models.However,the classical animal behavior analysis method relies on the artificial definition of metrics,and uses oversimplified features such as trajectories and velocities of animal to quantify behavior,which cause the questions of subjective description and information loss.Thus,an accurate behavioral tracking and analysis method without human supervision is urgently needed.Recently,the development of artificial intelligence provides a chance for objective and precise depiction of structured patterns of animal behavior.In this study,we developed and applied related algorithm to analyze animal behavior subtly based on artificial intelligence and machine learning technologies.Firstly,focusing on the dynamic,parallel and hierarchical characters of animal behavior,the deep learning poses estimation method and unsupervised machine learning algorithm were used to decompose behavior into subtle segment modules in mice.Then,these behavioral modules were represented in low dimensional space by unify metric.Based on the algorithm,the decomposition of looming stimuli induced innate behavior and controlled hunting behavior was demonstrated.The deep learning model was used to extract the inner sematic features of them.Next,tensor principle analysis and Markov process were used to build up mathematic model of their inner behavior states.At last,combining with fast high-resolution miniature two-photon microscopy and independently developed three dimensional behavior capturing system,the spontaneous neural activity decoding by correlated behavior was achieved.This result demonstrated the integration of subtle behavior data and corresponding neural activities.In summary,this study comprehensively investigated the structure features of animal behavior.Based on these features,the artificial intelligence and other algorithms were used to analyze animal behavior.The integrated algorithm was applied in the analysis of innate behavior and spontaneous neural-behavioral correlations.This study could provide subtle behavioral data support and reference for other related studies.