Research map / Probabilistic and scientific modeling / Dynamical systems
NeuroFlow: Toward Unified Visual Encoding and Decoding from Neural Activity
Weijian Mai, Mu Nan, Yu Zhu, Jiahang Cao, Rui Zhang, Yuqin Dai, Chunfeng Song, Andrew Luo, Jiamin Wu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add compact latent z_c (ablation) | NeuroFlow · NSD (Subject 1) | Encoding-to-decoding Inception: 97.2% -> 98.5% | Improved |
| add variational sampling (KL term) (ablation) | NeuroFlow · NSD (Subject 1) | Synthetic retrieval: 11.5% -> 96.4% | Improved |
| add variational sampling (KL term) (ablation) | NeuroFlow · NSD (Subject 1) | Raw retrieval: 75.1% -> 86.4% | Improved |
| add cycle-consistency loss (ablation) | NeuroFlow · NSD (Subject 1) | Encoding-to-decoding CLIP: 95.5% -> 98.7% | Improved |
| add contrastive alignment losses (L_clip, L_cyc) (ablation) | NeuroFlow · NSD (Subject 1) | Raw retrieval: 0.3% -> 86.4% | Improved |
| add contrastive alignment losses (L_clip, L_cyc) (ablation) | NeuroFlow · NSD (Subject 1) | Decoding Inception: 57.6% -> 95.9% | Improved |
| add cross-modal flow matching (XFM) (ablation) | NeuroFlow · NSD (Subject 1) | Encoding-to-decoding Inception: 60.3% -> 98.5% | Improved |
| replace XFM with MSE loss (ablation) | NeuroVAE · NSD (Subject 1) | Encoding-to-decoding Inception: 60.3% -> 51.0% | Hurt |
| replace XFM with two independent linear projections (ablation) | NeuroVAE · NSD (Subject 1) | Encoding-to-decoding Inception: 60.3% -> 90.3% | Improved |
| cosine instead of linear interpolation schedule (ablation) | NeuroFlow · NSD (Subject 1) | Decoding CLIP: 94.4% -> 95.0% | Improved |
| increase sampling steps from 20 to 30 (ablation) | NeuroFlow · NSD (Subject 1) | Decoding Inception: 95.9% -> 95.7% | No clear effect |
| increase sampling steps from 10 to 20 (ablation) | NeuroFlow · NSD (Subject 1) | Decoding Inception: 95.0% -> 95.9% | Improved |
Closest papers by meaning
- Brain decoding: toward real-time reconstruction of visual perception (ICLR 2024 · similarity 0.89)
- Flexible Context-Driven Sensory Processing in Dynamical Vision Models (NeurIPS 2024 · similarity 0.88)
- MoRE-Brain: Routed Mixture of Experts for Interpretable and Generalizable Cross-Subject fMRI Visual Decoding (NeurIPS 2025 · similarity 0.88)
- Neuro-Vision to Language: Enhancing Brain Recording-based Visual Reconstruction and Language Interaction (NeurIPS 2024 · similarity 0.88)
- MEIcoder: Decoding Visual Stimuli from Neural Activity by Leveraging Most Exciting Inputs (NeurIPS 2025 · similarity 0.88)
- Toward Generalizing Visual Brain Decoding to Unseen Subjects (ICLR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
