On January 12,the reporter learned from the Shenzhen Institute of Advanced Technology,Chinese Academy of Sciences that Wei Pengfei,a researcher of the Institute of Brain Cognition and Brain Disease,and his team applied AI technology to animal identification and neuroscience research,and proposed a small sample learning computing framework model to study social behavior.This model can solve multiple difficulties in accurately detecting animal social behavior and has the potential to innovate research paradigms in the neural loop mechanism of social behavior.The relevant research results are published in Nature Machine Intelligence.
  In recent years,the application of AI technology in traditional behavioral research has become increasingly widespread,and AI animal behavior tracking technologies such as DeepLabCut,SLEAP,MoSeq,etc.are becoming important research tools for neuroscientists.However,the above technologies still cannot achieve massive data annotation when analyzing multiple animal targets and animal free social behavior.In addition,the use of these technologies can also lead to low accuracy in identifying animal targets for continuous tracking.
  Based on this,the research team proposed a bidirectional transfer learning computational framework model.Using this model,researchers can achieve multi animal social identity recognition without the need to annotate animal identity data in advance.It is understood that the accuracy of this recognition exceeds 90%,which can fully meet the accuracy requirements of animal social experiments.
  "The design concept of the bidirectional transfer learning computer framework model is inspired by the working mechanism of the brain.In non social scenarios,distinguishing the identity of each animal is very simple.The animal identity information that these models already know can be transferred to multi animal social scenarios."Wei Pengfei said that this model solves the problem of AI needing to manually annotate a large amount of data to achieve multi animal identity recognition,Implemented zero sample multi animal social identity recognition.
  "Quantifying multi animal behavior is crucial for interpreting animal social behavior and has broad applications in neuroscience and ecology,"commented Trenton Jedd,senior editor of the journal Nature Machine Intelligence,on the study."In the future,AI enabled neuroscience research will provide guidance for implementing more precise and personalized non-invasive neural regulation,which is expected to help humans further understand complex mental disorders,"said Wei Pengfei.
  (Originally published in the 6th edition of Science and Technology Daily on January 15,2024)