Vision-language representations link food perception to subjective valuation

A multimodal vision-language deep neural network predicts subjective food value from food images more accurately than vision-only models, using human ratings (N = 194 and 200).

• Why it matters: It remains unclear how the brain computes subjective food values from visual input, despite these values guiding dietary choices.
• What they did: The study compared vision-language versus vision-only models on food images with human ratings (N = 194 and 200), then used representational analyses and fMRI multivariate mapping (N = 31).
• The result: Vision-language features preferentially encoded higher-level attributes like perceived healthiness and less image-level detail, and showed greater correspondence than vision-only features in the fusiform gyrus.

AI summary of the abstract · may contain errors · not medical advice · read the paper.