CAREER: Neural Representations of Naturalistic Behavior: From Perception to Cognition to Action

Animals survive by making flexible decisions as they move through complex environments. They look around, gather information, change strategies, and act on internal goals, but it remains difficult to determine how the brain produces these behaviors. Modern recording technologies can now measure brain activity and behavior in great detail, creating an opportunity to study behavior in settings that are closer to everyday life than highly simplified laboratory tasks. This project will develop computational tools that connect natural behavior to brain activity and help answer not only what an animal did, but why it behaved that way. The work will advance basic understanding of how the brain supports flexible behavior and may improve future studies of neurological and psychiatric disorders, brain-machine interfaces, and artificial intelligence systems that adapt to changing conditions. The project will also create new educational opportunities through research-based courses, undergraduate participation in neuroscience and artificial intelligence research, mentoring for K-12 teachers and students, and broader access to interdisciplinary training. The research is based on the idea that natural behavior contains measurable evidence about the internal computations that support perception, decision-making, and action. It will pursue three independent and complementary threads. The first objective will discover sensory strategies from freely moving behavior by developing inverse reinforcement learning methods that infer goals and strategies from behavior and visual input. These methods will determine how visual scenes and recent behavioral history shape active sensing, reward seeking, and disengagement, and whether these strategies help explain activity in primary visual cortex. The second objective will focus on cognitive state modeling by developing neural-behavioral dynamical models. These models will test whether similar outward behaviors can arise from different internal cognitive processes when prefrontal cortex activity is analyzed together with behavior. The third objective will focus on motor and reward representation by developing continuous, compositional motor-policy models together with distributional inverse reinforcement learning. These methods will study how motor cortex and striatum encode reusable movement patterns, task-specific combinations of those patterns, and variability in rewards. Together, these threads will test whether behavior data can reveal mechanistic structure across perception, cognition, and action, yielding circuit-level biological insight that descriptive behavior models cannot provide. The project will evaluate the methods on existing collaborator datasets and public neuroscience resources, release software and tutorials, and integrate the research with undergraduate research experiences, K-12 outreach, teacher mentoring, new courses, and interdisciplinary workshops. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2540614 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Anqi Wu | Institution: Georgia Tech Research Corporation, ATLANTA, GA | Award Amount: $354,774 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2540614 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2540614.html