emomind-affective-brain-decoding

name

emomind-affective-brain-decoding

description

End-to-end pipeline for decoding affective captions directly from fMRI signals using continuous emotion vectors and classifier-free guidance rewriting. Use when: brain-to-text with emotion/affect, fMRI affective decoding, continuous emotion representation from brain signals, individualized affective caption generation, classifier-free guidance for neural decoding. Activation: emomind, affective brain decoding, emotion fMRI, brain-to-text emotion, affective caption, continuous emotion vector, classifier-free guidance neural.

EmoMind: Decoding Affective Captions from Brain fMRI

First end-to-end pipeline for decoding affective captions directly from fMRI signals using continuous 34D emotion vectors, outperforming label-prompted GPT-4 across subject-specificity, structural geometry, and causal control axes.

Metadata

• Source: arXiv:2605.16739
• Authors: Bilal A. Mohammed, Lin Gu, Ruogo Fang
• Published: 2026-05-16

Core Problem

Brain-to-text systems recover semantic content but discard affect. Language models can generate emotional text from categorical labels, but labels collapse rich inter-subject variability into coarse discrete bins.

Key Innovation

Continuous emotion representation vs discrete labels:

• Decode a 34-dimensional continuous emotion vector from fMRI
• Avoids information loss from categorization
• Preserves individual affective structure

Methodology: Two-Stage Pipeline

Stage 1: Neutral Scene Description Retrieval

• Brain-decoded visual features → semantically grounded neutral scene description
• Standard brain-to-text for content (no affect)

Stage 2: Affective Rewriting with Classifier-Free Guidance

• Rewrite neutral description using decoded continuous emotion vector