Shogo Ohmae: Predictive world models in the neocortex and the cerebellum скачать в хорошем качестве

Shogo Ohmae: Predictive world models in the neocortex and the cerebellum

The neocortex and the cerebellum play important roles in a variety of cognitive processes. The cerebellum and neocortex have almost uniform circuit architectures across all functional domains, suggesting the existence of versatile and general-purpose computational mechanisms. We comprehensively compared the neocortex, cerebellum, and AI across diverse functional domains. Our theoretical analysis revealed striking convergent evolution between AI and brain processing. These parallels suggest that the neocortex and cerebellum acquire world models that enable them to (1) predict (generate) future information, (2) understand abstract features of the world, and (3) generate appropriate outputs, in a unified manner—foundational processes underlying diverse functions (Ohmae S & Ohmae K, 2024, arXiv). Next, we examined the hypothesis that the cerebellum can predict and model neocortical processing. This hypothesis is grounded in the anatomical observation that the inferior olive—the origin of prediction-error signals crucial for cerebellar learning —receives substantial disynaptic projections from the neocortex. Accordingly, it posits that the cerebellum learns to predict neocortical signals through this neocortico-olivo-cerebellar pathway (Ito, 2008; Inoue et al., 2016; Sokolov et al., 2017). To test this, we recorded cerebellar activity when a predictable neocortical signal after a visual stimulus was evoked. We found that Purkinje cells (1) receive this neocortical signal via the neocortico-olivo-cerebellar pathway as a teaching signal, and (2) learn to predict this neocortical signal. Thus, the cerebellum constructs predictive internal models not only of the external world but also of neocortical processing. Shogo Ohmae earned an M.D. from Kyoto University and then a Ph.D. from Juntendo University. He conducted postdoctoral research at Hokkaido University, where he was mentored by Masaki Tanaka, the University of Pennsylvania, and then Baylor College of Medicine, where he was mentored by Javier Medina. Subsequently he was appointed Assistant Professor at Baylor College of Medicine, where he received an NIH R34 grant. He is currently Assistant Investigator at the Chinese Institute for Brain Research, Beijing, where he investigates neocortical-cerebellar cognitive functions such as time and language processing, using large-scale neuronal recordings and AI-based modeling.