StimFlow: Synthesizing Brain Activity from Images via Stimulus-conditioned Flow Matching
We propose StimFlow, which synthesizes fMRI directly in native voxel space using conditional flow matching with a linear path.
ProofPaper ↗
Key points
- The inherent variability of fMRI responses across trials encourages a generative view of the stimulus-to-response mapping.
- Prior generative encoders often operate in compressed latent spaces and condition predominantly on semantic features, which provide limited low-level cues for early visual cortex.
- To match the visual hierarchy, StimFlow combines structural DINOv2 features with semantic CLIP features, which prove complementary along the cortical processing stream.
- Generation in voxel space allows direct ROI analysis and concept localization without a decoder.
Sources (1)
- [1]StimFlow: Synthesizing Brain Activity from Images via Stimulus-conditioned Flow MatchingarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 11, 04:45 AM
We propose StimFlow, which synthesizes fMRI directly in native voxel space using conditional flow matching with a linear path.
The inherent variability of fMRI responses across trials encourages a generative view of the stimulus-to-response mapping.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 10, 2026Anchors Enhance Controllability and Reasoning of Diffusion Language Models
- Oct 10, 2026Compositional Average Velocity Modeling for Efficient Diffusion Offline Reinforcement Learning
- Oct 10, 2026Verify, Act, Distill: Turning Test-Time Compute into Test-Time Adaptation for Vision-Language-Action Models
- Oct 9, 2026ConicFlow: constrained Inference-time optimal control for flow matching models
- Oct 9, 2026CorrFlow: Exact and Permutation-Equivariant Flow Matching on the Correlation Elliptope
- Oct 9, 2026BigCell: Generating Gigapixel Whole-Slide Images