Researchers from Carnegie Mellon University, Bonn University Hospital, and the University of Bonn have created an open-source platform known as A-SOiD that can learn and predict user-defined behaviors from videos. The research results have now been published in the journal Nature Methods.
This technique is very effective in learning to classify various animal and human behaviors. This will affect not just behavior, but stock markets, earthquakes, proteomics, and anything else where there is an identifiable pattern. This is a powerful pattern recognition machine. ”
Eric Ittori, Eberly Family Associate Professor of Biological Sciences, Carnegie Mellon University
Unlike many artificial intelligence (AI) programs, A-SOiD is not a black box. Instead, the researchers allowed the program to relearn what went wrong. They first trained the program using a portion of the dataset, focusing on the program's weak beliefs. When a program is uncertain, the algorithm strengthens its beliefs in its training data.
A-SOiD has been taught to focus on algorithmic uncertainty rather than treating all data the same, says Alex Hsu, a recent Ph.D. . The Carnegie Mellon University graduate said it avoids common biases found in other of his AI models.
AI tools process all classes in the dataset appropriately
“This is another way to enter data,” Sue said. “Typically, people look at entire datasets for whatever behavior they're looking for, with little understanding that the data can be imbalanced, meaning that some behaviors are well-represented in the set, and others are well-represented in the set. , meaning that there may be behaviors that are not well represented. “This bias can propagate from the prediction process to the experimental results. Our algorithm only By learning, we balance the data; our method is good at fairly representing all classes in the dataset.”
A-SOiD is trained in a supervised manner, making it highly accurate. Given a data set, it is possible to determine the difference between a person's normal tremor and the tremor of a person with Parkinson's disease. It also serves as a complementary technique to his B-SOiD, an unsupervised behavioral segmentation platform released two years ago.
In addition to being an effective program, A-SOiD is highly accessible, runs on regular computers, and is available as open source on GitHub.
A-SOiD is available to everyone involved in science
Jens Tillmann, a postdoctoral researcher at Bonn University Hospital, Bonn, said part of the impact was the idea of opening the program to all researchers.
“This project would not have been possible without the open science mindset that both of our labs, as well as the entire neurobehavioral community, have demonstrated in recent years,” Professor Tillman said. “I am excited to be part of this community and look forward to future collaborative projects with other experts in this field.”
Ittli and Martin K. Schwartz, principal investigator at the University Hospital Bonn and member of the University of Bonn's Interdisciplinary Research Area (TRA) Life and Health, are conducting their research to further investigate this relationship. I am planning to use A-SOiD in my room. between the brain and behavior. Yttri plans to use A-SOiD in combination with other tools to investigate the neural mechanisms underlying spontaneous behavior. Schwartz uses A-SOiD in combination with other behavioral modalities for fine-grained analysis of known behaviors in social interactions.
Both Ittori and Schwartz said they hope A-SOiD will be used by other researchers across disciplines and countries.
“A-SOiD is an important development that enables AI-based entry into behavioral classification, and therefore represents an excellent and unique opportunity to better understand the causal relationships between brain activity and behavior.” Schwartz said. “We also hope that the development of A-SOiD will serve as an effective catalyst for future collaborative research projects focused on behavioral research not only in Europe but also across the Atlantic.”
sauce:
Reference magazines:
Tillman, J.F.; other. (2024). A-SOiD is an active learning platform for data-efficient behavioral discovery under expert guidance. nature method. doi.org/10.1038/s41592-024-02200-1.