Speaker: Isıl Poyraz Bilgin, Ph.D., Postdoctoral Fellow, IUGM Research Center and Carnegie Mellon University (USA)
Summary :
Recent advances in artificial intelligence have brought language models (LMs) closer than ever to capturing certain facets of human cognition. Yet, despite their remarkable linguistic abilities, these models remain fundamentally grounded in textual patterns and lack the rich, multimodal grounding that shapes human thought and perception. In this talk, I will present our recent work on brain-informed language modeling, in which we developed a novel brain–AI alignment framework that trains large-scale language models directly on brain activity recorded during naturalistic experiences. Leveraging the vast CNeuroMod dataset, comprising tens of hours of fMRI recordings collected while participants watched richly contextual audio-visual naturalistic stimuli (e.g., movies, narratives, books), we align linguistic representations with neural dynamics to investigate how language models can internalize human-like cognitive structure. By aligning linguistic representations with neural dynamics, we demonstrate that brain-guided fine-tuning enhances these models’ ability to predict brain responses, generalize across individuals and stimuli, and internalize multimodal structure from neural signals. Beyond improving encoding accuracy, this approach enables language models to better capture and utilize the sensory, social, and contextual richness inherent in human cognition. Finally, I will discuss how brain-aligned models provide a new framework for studying the neural basis of meaning and may ultimately inform personalized, cognitively grounded AI systems for understanding brain function across the lifespan.
Biography :
Dr. Isıl Poyraz Bilgin is a Postdoctoral Research Fellow jointly affiliated with Carnegie Mellon University (USA) and the CRIUGM (Courtois NeuroMod). She specializes in advancing the emerging field of NeuroAI, which leverages neural data and cognitive principles to guide the development of AI systems grounded in the architecture and dynamics of the human brain. By combining fMRI and EEG, she designs multimodal modeling pipelines that capture the spatio-temporal complexity of human cognition and translate these dynamics into cognitively enriched models of the brain. Dr. Bilgin is a dedicated advocate for open, transparent, and inclusive science. She has served as Director of the international BrainHack School, conducted across nine global hubs, and as a teaching assistant, mentoring students in the effective implementation of open-science tools and practices in their research. Her teaching and mentorship experience spans from her early training in mathematics to her recent roles in the Montreal AI and Neuroscience (MAIN) conference educational series, where she contributed to lectures and hands-on tutorials in machine learning and NeuroAI. She has also mentored projects on ethics and interpretability of AI systems, through a collaborative program between the Turing Institute’s Open Life Science (UK) and the Digital Research Alliance of Canada. In addition, she co-leads and contributes to community-driven neuroinformatics and open-source initiatives, combining methodological innovation with a deep commitment to ethical, reproducible, and interdisciplinary research.