[Compressed Information Processing Team Seminar] Geometric Structure of Brain Representation Space and AI Latent Space: Determinants of Brain–AI Integration Performance

Prof. Takufumi Yanagisawa (Osaka University) presents recent work on zero-shot brain decoding that maps neural activity into AI latent spaces, and on closed-loop Brain–Machine Interfaces (BMI) that retrieve images from large databases via these mappings. The talk argues that the representation geometry of both the AI latent space and the brain's internal representations critically determines decoding accuracy and generalization in Brain–AI integration, and discusses medical applications — communication and motor function restoration for patients with severe motor impairment — via implantable devices that combine decoding, output generation, and external device control. Hybrid format: Zoom for all registrants; in-person at RIKEN AIP Nihonbashi Office (RIKEN affiliates only).