Large Language Models in Neuroscience

Abstract

The development of artificial intelligence (AI) systems, in particular, so-called foundation models and large language models, has opened a new era at the intersection of AI and neuroscience. These models allow working with diverse datasets across different modalities.

Compared to classical computational approaches, which mainly relied on traditional machine learning methods, these models represent a significant step forward. They demonstrate strong generalization capabilities and can capture complex spatio-temporal dependencies found in the data. This is achieved, in particular, through end-to-end learning directly on raw data. Foundation models can potentially be applied in all major neurobiological fields, encompassing neuroimaging and data processing, brain-computer interfaces and neural decoding, molecular neurobiology and genomic modeling, clinical care, and disease-specific applications, including neurological and psychiatric disorders. These models demonstrate the ability to solve fundamental computational neuroscience problems, including multimodal integration of neural data, spatio-temporal pattern interpretation, and the development of translational frameworks for clinical applications.