对应论文
Computational single-neuron mechanisms of visual object coding in the human temporal lobe
视频简介
This Nature Communications paper gives a mechanistic account of human visual object coding across ventral temporal cortex and medial temporal lobe. It combines population-level intracranial signals with single-neuron recordings to connect feature-axis representations with sparse higher-level object selectivity. The key contribution is a computational framework in which VTC neural axes form a feature space where objects cluster by perceptual and conceptual relationships, while MTL neurons encode receptive fields inside that feature space. This links dense visual representations to sparse memory-like object codes. For this repository, the value is NeuroAI framing: it offers a concrete bridge between feature geometry, category abstraction, and sparse high-level representations. That is directly relevant to multimodal representation learning, object encoding, and model-brain alignment work. It is not collected as a paradigm-level AI method because the contribution is primarily neuroscientific rather than a deployable model architecture. Its durable value is as a strong mechanistic reference for brain-inspired visual representation.