Research map / Label-efficient and robust learning / Self-supervised and contrastive
Process-conditioned Pretraining with Topographic Spatial Retrieval for Large EEG Models
Yi Ding, Muyun Jiang, Weibang Jiang, Shuailei Zhang, XINLIANG ZHOU, Chenyu Liu, Shanglin Li, Yong Li, Cuntai Guan
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes process-conditioned masked pretraining and topographic spatial retrieval for large EEG foundation models to exploit experimental mental-process cues and montage-independent spatial structure.
Problem
Large EEG foundation models pretrained on heterogeneous corpora underuse experimental mental-process cues and learn montage-dependent spatial representations, with spatial structure modeled only implicitly.
Approach
Conditions masked pretraining on mental-process cues from experimental protocols and adds a topographic spatial retrieval mechanism to make electrode spatial structure explicit and montage-agnostic, instead of a single general-purpose implicit representation.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add masked reconstruction pretraining objective (ablation) | BrainPro · FACED (9-class) | ACC-B: 0.5391 -> 0.5937 | Improved |
| Add separation loss between shared and process branches (ablation) | BrainPro · BCI-IV-2a (4-class) | ACC-B: 0.4870 -> 0.5674 | Improved |
| Use topographic retrieval instead of random retrieval (ablation) | BrainPro · BCI-IV-2a (4-class) | ACC-B: 0.5125 -> 0.5674 | Improved |
| Add process-guided spatial prior to reconstruction weights (ablation) | BrainPro · BCI-IV-2a (4-class) | ACC-B: 0.5496 -> 0.5674 | Improved |
| Use correct protocol-derived process cues instead of random cues (ablation) | BrainPro · FACED (9-class) | ACC-B: 0.5269 -> 0.5937 | Improved |
| Use process-conditioned multi-branch encoders instead of matched-size single encoder (ablation) | BrainPro · FACED (9-class) | ACC-B: 0.5683 -> 0.5937 | Improved |
| Add large-scale self-supervised pretraining (ablation) | BrainPro · FACED (9-class) | ACC-B: 0.4244 -> 0.5937 | Improved |
| Add affective branch to shared encoder (ablation) | BrainPro · FACED (9-class) | ACC-B: 0.5635 -> 0.5937 | Improved |
| Add both affective and motor branches (S+A+M) vs S+A (ablation) | BrainPro · FACED (9-class) | ACC-B: 0.5937 -> 0.5988 | Improved |
| Add auxiliary residual branch to all encoders (S+A+M+O) (ablation) | BrainPro · FACED (9-class) | ACC-B: 0.5988 -> 0.5824 | Hurt |
| Use motor branch instead of affective branch on motor imagery (ablation) | BrainPro · BCI-IV-2a (4-class) | ACC-B: 0.5499 -> 0.5674 | Improved |
| Drop 50% of channels at inference | BrainPro vs LaBraM and CBraMod · FACED (9-class) | ACC-B | Improved |
Closest papers by meaning
- BaRISTA: Brain Scale Informed Spatiotemporal Representation of Human Intracranial Neural Activity (NeurIPS 2025 · similarity 0.88)
- EEGPT: Pretrained Transformer for Universal and Reliable Representation of EEG Signals (NeurIPS 2024 · similarity 0.87)
- Standing on the Shoulders of Giants: Rethinking EEG Foundation Model Pretraining via Multi-Teacher Distillation (NeurIPS 2026 · similarity 0.87)
- CSBrain: A Cross-scale Spatiotemporal Brain Foundation Model for EEG Decoding (NeurIPS 2025 · similarity 0.86)
- MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding (NeurIPS 2026 · similarity 0.86)
- Brain-OF: An Omnifunctional Foundation Model for fMRI, EEG and MEG (NeurIPS 2026 · similarity 0.86)
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.
