Research map / Image generation / Autoregressive and latent generation
EEGFaceSem: An EEG Benchmark with Paired Generative Latents for Semantic Visual Modeling
Jun Ma, Michiel Spape, Keith M Davis, Tuukka Ruotsalo
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.
Introduces EEGFaceSem, an EEG benchmark pairing face-stimulus EEG recordings with generative latent codes to enable controlled study of semantic visual modeling.
Problem
Modeling EEG neural responses to visual semantics is limited by reliance on static real-world image datasets that lack fine-grained, controllable variation of visual features.
Approach
A benchmark that pairs EEG responses with generative latents so that visual features can be controlled and systematically varied, enabling isolation of fine-grained feature changes and their neural relations.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| train on many subjects (cross-subject LOSO) instead of single-subject for EEGNet (ablation) | EEGNet · EEGFaceSem mean AUC | AUC: 0.7416 -> 0.7715 | Improved |
| train on many subjects (cross-subject LOSO) instead of single-subject for LDA (ablation) | LDA · EEGFaceSem mean AUC | AUC: 0.7206 -> 0.6033 | Hurt |
| fine-tune the cross-subject model on target subject data (subject-adapted) (ablation) | EEGNet · EEGFaceSem mean AUC | AUC: 0.7715 -> 0.8063 | Improved |
| fine-tune the cross-subject model on target subject data (ablation) | EEGPT · EEGFaceSem mean AUC | AUC: 0.7003 -> 0.6824 | Hurt |
| aggregate per-stimulus predictions across 30 subjects by majority voting (ablation) | EEGNet (LOSO) · EEGFaceSem mean AUC | AUC: 0.7715 -> 0.9539 | Improved |
| use frozen pretrained EEGPT foundation model instead of simple MLP/EEGNet | EEG classifiers · EEGFaceSem cross-subject | AUC: 0.7715 -> 0.7003 | Hurt |
| use previous-trial epoch instead of current epoch to predict current label (RSVP carry-over control) (ablation) | LDA · EEGFaceSem single-subject | AUC: 0.7206 -> 0.5825 | Hurt |
| truncate the epoch window to end before next stimulus onset (ablation) | LDA on X_t-1 · EEGFaceSem | AUC: 0.5825 -> 0.5242 | Hurt |
| block-wise held-out split instead of shuffled folds (ablation) | LDA · EEGFaceSem | AUC: 0.7206 -> 0.7225 | No clear effect |
| restrict to trials with non-target neighbors and truncated window (ablation) | LDA whole-block · EEGFaceSem | AUC: 0.7225 -> 0.6002 | Hurt |
| condition AAE on brain-decoded labels instead of random labels (ablation) | AAE image generation · blinded paired annotation | correct-selection rate: 0.516 -> 0.747 | Improved |
| condition AAE on explicit labels instead of brain labels (ablation) | AAE image generation · blinded paired annotation | correct-selection rate: 0.747 -> 0.825 | No clear effect |
Closest papers by meaning
- MindAlign: Bridging EEG, Vision, and Language for Zero-Shot Visual Decoding (NeurIPS 2026 · similarity 0.91)
- EEG2Video: Towards Decoding Dynamic Visual Perception from EEG Signals (NeurIPS 2024 · similarity 0.91)
- What Does the Brain See? Multiview Neural Representation to Demystify the Brain-Visual Alignment (NeurIPS 2026 · similarity 0.91)
- BrainOmni: A Brain Foundation Model for Unified EEG and MEG Signals (NeurIPS 2025 · similarity 0.90)
- Decoding Natural Images from EEG for Object Recognition (ICLR 2024 · similarity 0.89)
- Visual Decoding and Reconstruction via EEG Embeddings with Guided Diffusion (NeurIPS 2024 · similarity 0.89)
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.
