Researchers at EPFL have developed TopoLM, an AI language model that mirrors the brain's language processing by capturing both the function and spatial arrangement of neurons. This model replicates the functional grouping of neurons and their spatial organization within the brain's cortex.
Unlike previous AI models that focused on individual clusters of functional neurons, TopoLM predicts how the brain's language system develops its spatio-functional organization. Professor Martin Schrimpf explains that TopoLM develops spatial clusters of internal components that functionally correspond to activity observed in the human brain during language processing. The model suggests that a basic rule governs spatial clusters in the brain, where nearby neurons behave similarly.
TopoLM offers a framework for enhancing the functional alignment of AI with human cognition, with potential applications in brain-inspired computing and neurolinguistics. This research, presented at the International Conference on Learning Representations (ICLR) 2025, marks a step toward AI systems that are organized more like the human brain. The researchers plan to test the model's predictions in the human brain through imaging studies.