Research map / Vision-language and multimodal / Cross-modal learning
SATTC: Structure-Aware Label-Free Test-Time Calibration for Cross-Subject EEG-to-Image Retrieval
Qunjie Huang, Weina Zhu
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 subject-adaptive whitening (SAW) to EEG embeddings (ablation) | Standardized baseline · THINGS-EEG 200-way LOSO | Top-5 / Top-1 accuracy (%): Top-5 30.5 -> 36.4; Top-1 9.2 -> 13.7 | Improved |
| add candidate whitening on top of SAW (ablation) | +SAW · THINGS-EEG 200-way LOSO | Top-5 / Top-1 accuracy (%): Top-5 36.4 -> 36.8; Top-1 13.7 -> 13.5 | No clear effect |
| add fixed-k CSLS (k=12) on SAW+CW (ablation) | +SAW+CW · THINGS-EEG 200-way LOSO | Top-5 accuracy (%): 36.8 -> 38.1 | Improved |
| replace fixed-k CSLS with adaptive CSLS (ablation) | +SAW+CW+CSLS · THINGS-EEG 200-way LOSO | Top-5 / Top-1 accuracy (%): Top-5 38.1 -> 38.8; Top-1 14.1 -> 13.9 | Mixed |
| add structural expert with PoE fusion on adaptive CSLS (ablation) | +SAW+CW+Ada-CSLS · THINGS-EEG 200-way LOSO | Top-1 accuracy (%): 13.9 -> 14.8 | Improved |
| add structural expert with PoE fusion on adaptive CSLS (ablation) | +SAW+CW+Ada-CSLS · THINGS-EEG 200-way LOSO | Top-5 accuracy (%): 38.8 -> 38.4 | No clear effect |
| use standardized inference (cosine, l2, candidate whitening) instead of original ATM dot-product | ATM · THINGS-EEG 200-way LOSO | Top-5 / Top-1 accuracy (%): Top-5 20.0 -> 30.5; Top-1 5.5 -> 9.2 | Improved |
| apply SATTC calibration head to EEGNetV4 | EEGNetV4 · THINGS-EEG 200-way LOSO | Top-5 / Top-1 accuracy (%): Top-5 20.5 -> 34.8; Top-1 5.4 -> 10.8 | Improved |
| apply SATTC calibration head to EEGConformer | EEGConformer · THINGS-EEG 200-way LOSO | Top-5 / Top-1 accuracy (%): Top-5 11.6 -> 23.2; Top-1 2.5 -> 6.9 | Improved |
| apply SATTC calibration head to ShallowFBCSPNet | ShallowFBCSPNet · THINGS-EEG 200-way LOSO | Top-5 / Top-1 accuracy (%): Top-5 14.6 -> 30.8; Top-1 3.5 -> 11.1 | Improved |
| use adaptive CSLS instead of fixed-k CSLS for hubness (ablation) | CSLS · THINGS-EEG class popularity | hubness (class-occurrence curve) | Improved |
| use unlabeled calibration window of N=50 instead of N=200 (ablation) | SAW+CSLS · THINGS-EEG N-shot | fraction of upper bound: 94.8% of the N = 200 upper bound | No clear effect |
Closest papers by meaning
- MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding (NeurIPS 2026 · similarity 0.90)
- What Does the Brain See? Multiview Neural Representation to Demystify the Brain-Visual Alignment (NeurIPS 2026 · similarity 0.88)
- Visual Decoding and Reconstruction via EEG Embeddings with Guided Diffusion (NeurIPS 2024 · similarity 0.87)
- CSBrain: A Cross-scale Spatiotemporal Brain Foundation Model for EEG Decoding (NeurIPS 2025 · similarity 0.87)
- Decoding Natural Images from EEG for Object Recognition (ICLR 2024 · similarity 0.87)
- Neural‑Visual Decoding via Cognitive‑guided Adaptive Blurring and Information‑Constrained Alignment (NeurIPS 2026 · 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.
