AI model can reconstruct images from your brain (petapixel.com)

Researchers at the Weizmann Institute of Science have unveiled Brain-IT, an advanced AI model capable of reconstructing images from brain activity with impressive speed and accuracy. This breakthrough, developed by Prof. Michal Irani's team, significantly reduces the time required for image reconstruction from several hours to just about one. Unlike its predecessors, which often struggled with basic features like color and composition, Brain-IT excels in both content and detail, showcasing its potential for clearer, more precise image recreation. The model was trained on over 70,000 images viewed by eight participants while their brain activity was monitored. The researchers employed an innovative "encoder" that identifies shared patterns within brain scans across individuals, enabling predictions of brain activity from images. By dissecting each brain scan into 40,000 small sections and understanding the visual features associated with them, the team pinpointed 128 functional regions in the brain involved in image processing. This dual model configuration—an encoder for brain activity predictions and a decoder for image reconstruction—offers profound implications for not only enhancing communication for individuals with paralysis but also advancing our understanding of brain function in visual processing.