Mark Struman
Pronouns: He/Him
Research Mentor(s): Bing Ye
Research Mentor School/College/Department: Life Sciences Institute, University of Michigan / Medicine
Program:
Authors:
Session: Session 7: 4:40 pm – 5:30 pm
Poster: 11
Abstract
Behavioral scientists have long been plagued by the time-intensive efforts needed, in both field and laboratory, to make their observations. Now, in the advent of artificial intelligence technologies, both individual and dyadic behavior can be automatically recorded; however, within group settings, which introduce greater variability between an action’s actor and its target, classifying these interactions has posed difficulties. Within this study, using a deep-learning neural network with a ‘spotlight’ guided approach – treating each subject as its own main actor within a list of potentially observable behaviors – LabGym has been able to identify both individual and dyadic behaviors within a scalable setting. While other software-intelligent tools have been limited to interactions isolating just the two organisms interacting, the spotlight model has been observed extracting groups of dyadic and individual behavior patterns in group settings. Not only have these behavioral patterns been recognized, but LabGym has demonstrated the ability to map these behaviors with their respective pairs and individuals. Understanding these innovations, future behavioral studies may rely less on the constants of laboratories, but acquire meaningful observations within real-world applications as well. From a herd of cows grazing in their pasture to one primate interacting with an array of makeshift tools, LabGym is capable of finding pairs of interactions existing between all of these groups. Now, while identifying more abstract, group behavior still remains an issue, LabGym’s capabilities in taking the imperfect, bustling group interaction and being able to analyze the individualistic and dyadic behavior within it remains a strong step in the pursuance of AI technologies, and within expanding the utility of tools available to assist behavioral scientists.
