Get started
Work

EEG for Sony Award

Year: 2026
Type: other
Abstract: It has been long known in both neuroscience and artificial intelligence (AI) that "coupling" between neurons leads to a form of dynamically distributed system where the synchronization of neural oscillations emerges in self-organized spatiotemporal patterns, capable of representing more abstract concepts. By capitalizing on the unprecedented amount of human neuroimaging data, we hypothesize that a new understanding of neural mechanisms that govern the brain rhythms could inspire next-generation design principles for machine learning algorithms with greater efficiency and robustness. In this context, we aim to establish a NeuroAI framework that not only sheds new light on machine intelligence but also offers new insights into brain function, revolutionizing our understanding of both neuroscience and AI. Specifically, we aim to answer the following unexplored questions: (i) What is the governing equation underlying the dynamical system of oscillatory neural activities?} (ii) Is it possible to redefine the nature of intelligence by drawing inspiration from the dynamical system of neural oscillations? (iii) How can AI be designed to reason effectively and solve neuroscience problems without relying on ground-truth labels? We will investigate whether the neural mechanism of neural oscillations identified by BRICK aligns with existing findings in experimental neuroscience. The conceptualization is that the success of modeling brain rhythms via an artificial dynamic system of coupled oscillations gives rise to the "first principle" for emerging brain-inspired machine intelligence through the common mechanism of synchronization.Following this spirit, we will develop a set of biologically-inspired neural networks (BINNs) to reason and solve problems by regarding each data entity (such as graph node) as an oscillation and conducting \textbf{unsupervised learning via oscillatory synchronization}. Furthermore, we will enhance the scalability and generalizability up BINNs to real-world neuroimaging data by incorporating state-of-the-art machine learning techniques such as masked autoencoder (MAE) and large language model (LLM). Since the overarching goal of NeuroAI is to bridge the reciprocal relationship between neuroscience and AI, we will explore the challenging problem of whole-brain parcellation through cortical oscillations that emerge the individual's brain rhythm. We will evaluate the new functional parcellation approach on a large-scale human neuroimages and various clinical applications such as surgery planning and disease diagnosis.To promote the collaborative efforts of \framework{}, we will maintain and release the source code of BINNs to the neuroimaging and machine learning communities. The output of this project is a new interdisciplinary field that combines power of neuroscience and AI for scientific discovery.
DOI: 10.48321/d1d609c094
Author: Guorong Wu
Language: English
Open Access status: green
OpenAlex ID: W7204588066
Cites:
Cited by:
Related to: