English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Feynman-Style Breakdown: How the Brain 'Replays' During Mental Imagery — Science 2026 Paper on Shared Visual Codes

Forum topic · 小凯 · 2026-05-16

Summary

This English explainer distills a 2026 Science paper (DOI: 10.1126/science.adt8343) by Wadia, Rutishauser, and Tsao, in which the authors recorded 714 single neurons in the human ventral temporal cortex (VTC) of 16 epilepsy patients. Roughly 80% of visually selective neurons used an 'axis-coding' scheme, in which each cell encodes one linear projection direction in a 50-dimensional object-feature space derived from AlexNet's fc6 layer. During mental imagery in 6 patients, about 40% of axis-tuned neurons reactivated along the same axes, enabling linear decoders trained on perception data to read out imagined objects at 31.35% (versus a 16.67% chance baseline), and image reconstructions reached a median cosine similarity of 0.89. The findings support a 'replay hypothesis' over a 'simulation hypothesis' and offer single-neuron evidence that cortex runs a generative model, with implications for Alzheimer's, BCIs, and hallucination in psychosis.

This English explainer distills a 2026 Science paper (DOI: 10.1126/science.adt8343) by Wadia, Rutishauser, and Tsao, in which the authors recorded 714 single neurons in the human ventral temporal cortex (VTC) of 16 epilepsy patients. Roughly 80% of visually selective neurons used an 'axis-coding' scheme, in which each cell encodes one linear projection direction in a 50-dimensional object-feature space derived from AlexNet's fc6 layer. During mental imagery in 6 patients, about 40% of axis-tuned neurons reactivated along the same axes, enabling linear decoders trained on perception data to read out imagined objects at 31.35% (versus a 16.67% chance baseline), and image reconstructions reached a median cosine similarity of 0.89. The findings support a 'replay hypothesis' over a 'simulation hypothesis' and offer single-neuron evidence that cortex runs a generative model, with implications for Alzheimer's, BCIs, and hallucination in psychosis.

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177620085