INNS Webinar Series: NeuroAI

Описание: Abstract: NeuroAI creates synergies between the study of brains and AI advances because both of their neural architectures are massively parallel and highly connected by weights trained by learning algorithms. Horace Barlow suggested that the oriented filters in visual cortex, generally thought to be edge detectors, might be a compact way to represent natural scenes. We confirmed this hypothesis with Independent Component Analysis (ICA), an unsupervised learning algorithm, training on patches of natural scenes. The independent components were edge filters; each patch could be reconstructed from only a sparse set of components. When large-scale Convolutional Neural Networks (CNNs) with many processing layers became feasible a decade ago, they could recognize thousands of objects in images invariant to location, scale, and rotation. Similar progress has occurred in language processing, starting with NETtalk in the 1980s, which we trained to pronounce English text, a difficult problem in linguistics because of many irregularities. Today, Large Language Models (LLMs) talk to us on almost any topic in perfect syntax.

Speaker Bio: Terrence Sejnowski received a Ph.D. in Physics from Princeton University. He was a postdoctoral fellow at Princeton University and Harvard Medical School before being appointed to a faculty position in the Department of Biophysics at Johns Hopkins University in 1981. He moved to La Jolla in 1989 and is currently on the faculty of the Salk Institute for Biological Studies and the University of California at San Diego.

Sejnowski is a member of the National Academies of Sciences, Engineering, Medicine, and Inventors. He was also an Investigator with the Howard Hughes Medical Institute and was awarded the Brain Prize in 2024. He was elected to the Royal Society in 2025.

Sejnowski pioneered computational neuroscience and learning algorithms in neural network models. His research aims to understand the computational resources of brains and build linking principles from brains to behavior. He is a leader in NeuroAI, the recent convergence between neuroscience and AI. His laboratory developed Independent Component Analysis (ICA) for blind source separation, which is universally used for analyzing brain imaging using electroencephalography EEG and functional magnetic imaging (fMRI).