brAIn Seminar - Martin Schrimpf
November 5, 2026 3:30PM—4:30PM
Location:
Virtual Presentation - ET
-
Group Viewing and Remote Access - Zoom
Speaker:
MARTIN SCHRIMPF,
Assistant Professor and Head, NeuroAI LabSchool of Life Sciences and School of Computer and Communication Sciences, EPFLCo-founder and Chief Science Officer, Read/Write Labs
https://mschrimpf.altervista.org/
How can we understand the neural mechanisms underlying intelligence? I will argue that we are entering an era of digital brain models that are making great strides in forming an answer: these artificial neural networks not only reproduce human behaviors, but also do so via internally consistent representations. First, I will show that AI models optimized for real-world tasks, such as video understanding and language modeling, converge on strikingly brain-like solutions, which we quantify at scale through the Brain-Score community platform. But this convergence holds only up to a point: task optimization alone has begun to plateau, capturing roughly half of the explainable neural signal. I will make the case that the next leap will come from scaling neural data and optimizing models directly with large-scale brain recordings, which will lead to next-generation multimodal predictive brain models. Finally, I will show how the best models already let us move from understanding to intervention: guiding stimulation to induce visual percepts via topography, and designing NeuroAI-based treatments such as fonts that improve reading in dyslexia. I will close on where NeuroAI might go next.
—
Martin Schrimpf is a tenure-track Assistant Professor at EPFL, where he heads the NeuroAI Lab, and Co-Founder and Chief Scientist at Dandelion Science. His lab builds computational models of brain function, from sensory input to language and cognition. He then uses those foundation models as instruments: modeling dysfunction such as dyslexia and anhedonia, discovering treatment candidates, and guiding neuromodulation. Martin initiated and leads the community platform Brain-Score.org, with over 100 brain and behavior alignment benchmarks, and built state-of-the-art models including CORnet, TopoLM, and MIRAGE. He completed his PhD at MIT with Jim DiCarlo, and previously worked at Harvard, MetaMind/Salesforce, and Oracle, and co-founded two prior startups. His work has been featured by the BBC, Quanta, Science, and Scientific American, and recognized with a Takeda Fellowship, the Neuro–Irv and Helga Cooper Foundation Open Science Prize, and a Google.org Impact Challenge Prize. He is an AI2050 Early Career Fellow and an ELLIS Scholar.
Group Viewing (Location TBD) and Zoom Participation. See announcement.
For More Information:
rdkeller@cs.cmu.edu