Research map / Vision-language and multimodal / Cross-modal learning
CogMoE: Signal-Quality–Guided Multimodal MoE for Cognitive Load Prediction
Aamir Bader Shah, Yu Wen, Renjie Hu, Jiefu Chen, Jose L Contreras-Vidal, Xuqing Wu, Xin Fu
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 time-frequency synchronization and multimodal recovery pre-processing (ablation) | CogMoE · CL-Drive | accuracy / F1: +10.3% accuracy, +11.47% F1 over raw input | Improved |
| add MoE and CORTEX loss over a simple FFN (ablation) | CogMoE · CL-Drive | accuracy / F1: over 11% higher accuracy and F1 than the FFN baseline | Improved |
| replace single FFN (CogBasic) with quality-aware experts (ablation) | CogMoE · CL-Drive (Table 8) | accuracy / F1: 6–8% on both accuracy and F1 | Improved |
| route experts by signal quality instead of modality identity (ablation) | CogMoE · CL-Drive (4 modalities) | accuracy: 89.8 -> 94.5 | Improved |
| add Noise Resilient Expert (ablation) | CogMoE · CL-Drive | accuracy / F1: 2.8% accuracy drop, 2.0% F1 drop when removed | Improved |
| add Contextual Refinement Expert (ablation) | CogMoE · CL-Drive | accuracy / F1: 1.9% accuracy drop, 1.3% F1 drop when removed | Improved |
| use all three experts instead of only the High-Fidelity Expert (ablation) | CogMoE · CL-Drive | accuracy / F1: 9.2% accuracy drop, 9.6% F1 drop with HFE only | Improved |
| add noise suppression loss (ablation) | CORTEX loss · CL-Drive | accuracy: -1.2% accuracy and +105.7% noise error when removed | Improved |
| add refinement loss (ablation) | CORTEX loss · CL-Drive | accuracy: -1.5% accuracy when removed | Improved |
| add gating regularizer Rgate (ablation) | CORTEX loss · CL-Drive | accuracy: -7.3% accuracy; HFE handles 81% of samples when removed | Improved |
| add data augmentation (noise, shifts, channel dropout) | CogMoE · CL-Drive (4 modalities) | accuracy: +3.04% accuracy, +2.84% F1 | Improved |
| add data augmentation (noise, shifts, channel dropout) | RF · CL-Drive (4 modalities) | accuracy: 81.71 -> 79.34 | Hurt |
Closest papers by meaning
- EEG Benchmarking Needs a Task Specification Layer: NeuroDoc for Rulebook-Guided, Executable Benchmark Construction (NeurIPS 2026 · similarity 0.85)
- CORTEG: Foundation Models Enable Cross-Modality Representation Transfer from Scalp to Intracranial Brain Recordings (NeurIPS 2026 · similarity 0.84)
- NeuroAtlas: Benchmarking Foundation Models for Clinical EEG and Brain-Computer Interfaces (NeurIPS 2026 · similarity 0.83)
- What do EEG Foundation Models Capture from Human Brain Signals? (NeurIPS 2026 · similarity 0.83)
- CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding (ICLR 2025 · similarity 0.83)
- REVE: A Foundation Model for EEG - Adapting to Any Setup with Large-Scale Pretraining on 25,000 Subjects (NeurIPS 2025 · similarity 0.82)
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.
