It has become increasingly common to use representations extracted from modern AI models for language and vision to study these same processes in the human brain. This approach often achieves accurate prediction of brain activity, often accounting for almost all the variance in the recordings that is not attributable to noise. However, better prediction performance doesn't always lead to better scientific interpretability. This talk presents some approaches for the difficult problem of making scientific inferences about how the brain represents high-level meaning. We also discuss how to go beyond aligning AI representations and brains. Instead, we directly learn the representations used in a brain region from its activity recordings. Using modern AI tools, data from naturalistic neuroimaging experiments and other large scale datasets, we reconstruct the representations and preferences of individual voxels and suggest new subdivisions that are more refined than existing regions of interest. This perspective draws a close connection between brains and AI models, reveals new aspects of brain function, and can serve as the basis for more powerful brain computer interfaces.

Leila Wehbe is an associate professor in the Machine Learning Department and the Neuroscience Institute at Carnegie Mellon University. Her work is at the interface of cognitive neuroscience and computer science. It combines naturalistic functional imaging with machine learning both to improve our understanding of the brain and to find insight to build better artificial systems. She is the recipient of an NSF CAREER award, a Google faculty research award and an NIH CRCNS R01. Previously, she was a postdoctoral researcher at UC Berkeley and obtained her PhD from Carnegie Mellon University.

Cerebrovascular diseases, including stroke and related conditions, are among the leading causes of global morbidity and mortality. Effective clinical management of these complex conditions requires accurate risk assessment, timely intervention, and individualized treatment strategies. Cerebrovascular disorder demands innovative approaches for accurate risk prediction and effective clinical management. This talk presents NeuroAssist, a framework combining multimodal neurosymbolic AI with federated learning to address the challenges of hemorrhagic and ischemic stroke management.  It will explain how to perform privacy-preserving machine learning across institutions with human-understandable reasoning, to enhance trust and usability in clinical settings.  Through case study of cerebral aneurysm, we will explore how to empower clinicians with trustworthy AI tools to improve outcomes for patients with cerebrovascular diseases. Using cerebral aneurysm risk prediction as a case study, the talk will demonstrate how NeuroAssist empowers Neurosurgeons with AI-driven tools for subarachnoid hemorrhage prediction, stroke risk stratification, and personalized treatment optimization.

Dr. Khalid Malik is a Professor of computer science and director of cybersecurity at the College of Innovation and Technology, University of Michigan-Flint. His research centers on designing secure, intelligent, and decentralized decision support systems using multimodal, federated, trustworthy, and neuro-symbolic AI. In healthcare, he specializes in predicting cerebrovascular and cardiovascular events through clinical text and multiple medical imaging modalities (e.g., DSA, MRA). In cybersecurity, his research is directed towards developing forensic examiners to ensure the authenticity, integrity, and veracity of multimedia (audios, videos, images) and implementing web filtering using multimodal and neuro-symbolic AI. Dr. Malik’s research is funded by multiple National Science Foundation awards, the Brain Aneurysm Foundation, the Department of Energy, the Michigan Translational Research and Commercialization (MTRAC) Innovation Hub, MTRAC Life Sciences, and several national and international industry partners. He is a recipient of numerous accolades, not limited to Oakland’s Young Investigator Research Award (2018), SECS Outstanding Research Award (2019), and Distinguished Associate Professor Award (2021).

Human behavior is based on a complex interaction between perception, stored knowledge, and continuous evaluation of the world relative to plans and goals. Even simple tasks involve processes whose underlying circuitry is broadly distributed across the brain. A key component of this system is the Distributed Conceptual Network (DCN), which integrates perceptual information with memory in the service of current plans and goals, supporting attention, working memory and conscious experience. In this talk I will contrast the architecture and function of the DCN with current AI systems such as transformer-based LLMs and reinforcement learning agents.

Jack Gallant is co-Director of the Henry H. Wheeler Jr. Brain Imaging Center and the Class of 1940 Chair at the University of California at Berkeley. He holds appointments in the Departments of Neuroscience and Electrical Engineering and Computer Science, and is a member of the programs in Bioengineering, Biophysics, and Vision Science. He is a senior member of the IEEE, and served as the 2022 Chair of the IEEE Brain Community. Professor Gallant's research focuses on high-resolution functional mapping and quantitative computational modeling of human brain networks. His lab has created the most detailed current functional maps of human brain networks mediating vision, language comprehension and navigation, and they have used these maps to decode and reconstruct perceptual experiences directly from brain activity. Further information about ongoing work in the Gallant lab, links to talks and papers and links to online interactive brain viewers can be found at http://gallantlab.org.

TBD

Professor Sheth is an educator, researcher, and entrepreneur. He is the founding director university-wide AI Institute at the University of South Carolina. He is a Fellow of the IEEE, AAAI, AAAS and ACM, elected for his pioneering and enduring contributions to information integration,  distributed workflow processes, semantics, knowledge-enhanced computing, etc. He has (co-)founded four companies, three of them by licensing his university research outcomes, including the first Semantic Search company in 1999 that pioneered technology similar to what is found today in Google Semantic Search and Knowledge Graph. He is particularly proud of the success of his 45 Ph.D. advisees and postdocs in academia, industry research, and entrepreneurs.  He received his B.E. (Hons) from BITS-Pilani, India, and MS and PhD from the Ohio State University, USA. 

Dynamic causal modeling has been an influential approach for studying effective connectivity in the brain, as supported by about 1,400 papers on Pubmed mentioning this approach (as of January 2025). However, we can conceptualize this approach within the more comprehensive framework of model-driven analysis. In this framework, a biologically relevant generative model is designed and fitted onto experimental data to gain insight into potentially latent variables and processes. In this talk, I will present this general framework, connect it to existing siloed approaches, emphasize its crux (i.e., the inference of dynamical systems parameters), and discuss how AI and, more specifically, deep learning can help address this thorny problem. In doing so, I will also discuss how biological neural processes are modeled, one of the core elements of this approach.

Christian O’Reilly received his B.Ing(electrical eng.; 2007), his M.Sc.A. (biomedical eng.; 2011), and his Ph.D. (biomedical eng.; 2012) from the École Polytechnique de Montréal where he applied pattern recognition and machine learning to predict brain stroke risks. He was later a postdoctoral fellow at the Université de Montréal (2012-2014) and then an NSERC postdoctoral fellow at McGill's Brain Imaging Center (2014-2015) where he worked on characterizing EEG sleep transients, their sources, and their functional connectivity. From 2015 to 2018, he led the large-scale biophysically-detailed modeling of the thalamocortical loop at the Blue Brain project. He then worked as a Research Associate at the Azrieli Centre for Autism Research (McGill) where he studied brain connectivity in autism and related neurodevelopmental disorders. Since 2021, Christian joined the Department of Computer Science and Engineering, the Artificial Intelligence Institute (AIISC), and the Carolina Autism and Neurodevelopmental (CAN) Research Center at the University of South Carolina as an assistant professor in neuroscience and artificial intelligence.

Large language models (LLMs) can be powerful tools for data analysis in neuroscience. Their use as models of cognitive systems, however, is limited. I will present two examples such use. The first is related to the nature of semantic representations in the brain. By comparing the brain-LLM correspondence for models with and without vision, we can shed light on a central question in cognitive neuroscience of language: whether neural representations of concepts in semantic hubs are multimodal or amodal. In the second case, we lesion LLMs in the context of a picture naming task, a common benchmark for aphasia. I demonstrate that by varying lesioning parameters, we can obtain various types of naming errors seen in persons with aphasia. Moreover, we can replicate the profile of the type and proportion of naming responses seen in individual stroke survivors in many, but not all, cases. Such models of individual stroke survivors may be used in the future to efficiently test rehabilitation strategies. These results suggest a rich potential for LLMs as models of healthy and impaired cognitive systems.

Rutvik H. Desai is a Professor of Psychology at the University of South Carolina. His early career focused on artificial intelligence and machine learning, where he developed methods of pattern classification and optimization. During graduate training in computer science and cognitive science at Indiana University, his work focused on developing connectionist models of language acquisition, explaining how pattern-based language learning in children shifts to word-based learning. As a postdoctoral fellow, he studied neuroimaging under the mentorship of Jeffery Binder at the Medical College of Wisconsin, focusing on brain basis of language and semantics. Later he became a faculty member at the Medical College in Neurology, and studied morphology, reading, and representation of action concepts in the brain using neuroimaging and behavioral studies of patients with Parkinson’s. After ten years at the Medical College, he moved to the University of South Carolina where he directs a research lab. His lab studies language and especially representation of concepts and word meanings in the brain using a variety of methods. His work has shown how sensory-motor and conceptual systems of the brain are closely tied. His lab uses functional MRI, transcranial magnetic and direct current stimulation, and behavioral studies of healthy populations as well as those of patients with stroke or Parkinson’s to understand the neural basis of language processing. As part of the Center for the Study of Aphasia Recovery, he hopes to apply a variety of neuroimaging techniques to better understand, and develop more informed therapies for, aphasia resulting from stroke. His research is supported by the National Institute of Deafness and Communication Disorders.