Research map / Architectures and efficiency / Attention and transformers
SwiFT: Swin 4D fMRI Transformer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use absolute positional embedding instead of relative positional bias (ablation) | SwiFT · HCP sex classification | accuracy: 89.8 -> 92.9 | Helped |
| use absolute positional embedding instead of relative positional bias (ablation) | SwiFT · UKB age | MSE: 19.4 -> 18.2 | Helped |
| use absolute positional embedding instead of relative positional bias (ablation) | SwiFT · ABCD sex classification | accuracy: 80.2 -> 79.3 | Hurt |
| use absolute positional embedding instead of relative positional bias (ablation) | SwiFT · Efficiency | throughput (samples/sec): 94.17 -> 104.16 | Helped |
| use end-to-end 4D Swin instead of two-step TFF | TFF · HCP age | MSE: 13.8 -> 8.6 | Helped |
| use end-to-end 4D Swin instead of ROI-based BrainNetCNN | BrainNetCNN · ABCD sex classification | accuracy: 80.1 -> 79.3 | No clear effect |
| use end-to-end SwiFT instead of TFF | TFF · UKB intelligence | MSE: 0.997 -> 0.992 | No clear effect |
| SwiFT in place of TFF for efficiency | TFF · Efficiency | throughput (samples/sec): 53.60 -> 104.46 | Helped |
| contrastive pre-training on UKB before fine-tuning (ablation) | SwiFT · HCP/ABCD intelligence | MSE | Mixed |
| increase input time frames from 4/8 to 16/32 (ablation) | SwiFT · HCP and UKB intelligence | MAE | Helped |
| increase input time frames to 16/32 (ablation) | SwiFT · UKB age | MAE | Hurt |
Closest papers by meaning
- FAST-Brain: A Flow-Aligned Spatio-Temporal Surrogate Brain Model (NeurIPS 2026 · similarity 0.87)
- Brain Harmony: A Multimodal Foundation Model Unifying Morphology and Function into 1D Tokens (NeurIPS 2025 · similarity 0.86)
- Brain-OF: An Omnifunctional Foundation Model for fMRI, EEG and MEG (NeurIPS 2026 · similarity 0.86)
- Meta-Learning an In-Context Transformer Model of Human Higher Visual Cortex (NeurIPS 2025 · similarity 0.86)
- BrainWorld: A Structural-Prior-Conditioned Generative Model for Whole-Brain 4D fMRI Dynamics (NeurIPS 2026 · similarity 0.86)
- BaRISTA: Brain Scale Informed Spatiotemporal Representation of Human Intracranial Neural Activity (NeurIPS 2025 · similarity 0.85)
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.