51. New Approaches to Studying Rodent Behavior Using Deep Machine Learning
- Author
-
Ilya Kuznetsov, Ilya Kovalenko, Alexander Andreev, Danila Apushkin, Vyacheslav Kalchenko, Eugenia Ahremenko, and Eduard Korkotian
- Subjects
Correctness ,Artificial neural network ,business.industry ,Computer science ,Orange (software) ,Behavioral pattern ,Flexibility (personality) ,Machine learning ,computer.software_genre ,Field (computer science) ,Resource (project management) ,Artificial intelligence ,Transfer of learning ,business ,computer - Abstract
One of the fundamental tasks in behavioral neurophysiology is the study of movement control. During past decades, methodological arsenal in this field was limited to the expert's visual observation, especially in the open field paradigm when animals can move, groom, rare or explore objects. In the past few years, the massive adoption of machine learning and artificial intelligence has led to revolutionary changes in the field. Here we present a new and user friendly approach to the analysis of the behavior of rodents, based on a pre-trained and continuing to learn neural network, which is capable of exhibiting a high degree of flexibility and good level of performance. Open field behavior of two groups (juvenile and adult) female wistar rats has been recorded. Human expert identified several behavioral patterns for farther deep machine learning, based on open source Orange SqueezeNet, a pre-trained neural network. In addition, we used the transfer learning technique with three more layers over SqueezeNet with 10-fold cross validation. The correctness of the assessments, made by the neural network consisted 92 ± 3% based on unbiased and double-blinded expert evaluation. Moreover, younger and older rat groups revealed significant difference in such behavioral patterns as moving and hole testing, identified by the network. We conclude that our system is demonstrative, does not require high programming skills, looks intuitive, easy to use, is based on an open source resource, therefore, can be recommended as behavioral research tool and an educational platform.
- Published
- 2021
- Full Text
- View/download PDF