AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

ViEEG: Hierarchical Visual Neural Representation for EEG Brain Decoding

Minxu Liu, Donghai Guan, Chuhang Zheng, Chunwei Tian, Jie Wen, Qi Zhu

ICML 2026 · Vision-language and multimodal › Human and embodied modeling · approach: fMRI visual decoding · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add cross-attention hierarchical integration across the three EEG streams (ablation)ViEEG · THINGS-EEG subject-dependent top-1top-1 accuracy (%): 38.6 -> 40.9Improved
add cross-attention hierarchical integration across the three EEG streams (ablation)ViEEG · THINGS-EEG subject-independent top-1top-1 accuracy (%): 20.7 -> 22.9Improved
fuse contour, object and scene EEG embeddings instead of contour only (ablation)ViEEG · THINGS-EEG subject-dependent top-1top-1 accuracy (%): 14.8 -> 40.9Improved
use object or scene embedding alone instead of the contour embedding alone (ablation)ViEEG · THINGS-EEG subject-dependent top-1top-1 accuracy (%): 14.8 -> 28.9 (object), 28.6 (scene)Improved
use contour->object->context order rather than other permutations (ablation)ViEEG · THINGS-EEG subject-dependent top-1top-1 accuracy (%): best alternative 40.3 -> 40.9Improved
use ViEEG hierarchical encoder instead of NICE or ATM encoders (ablation)ViEEG · THINGS-EEG subject-dependent top-1top-1 accuracy (%): 37.8 (NICE) / 37.6 (ATM) -> 40.9Improved
use one cross-attention layer instead of none or deeper stacks (ablation)ViEEG · THINGS-EEGtop-1 accuracyImproved
use 3 attention heads instead of other head counts (ablation)ViEEG · THINGS-EEG subject-dependenttop-1 accuracyImproved
concatenate mask, object and scene embeddings (triple) instead of a single embedding for retrieval (ablation)ViEEG · THINGS-EEG zero-shot retrieval (Subject 8)retrieval accuracyImproved
replace NICE/ATM/CognitionCapturer with ViEEG for reconstructionSDXL-Turbo + IP-Adapter · THINGS-EEG reconstructionCLIP similarity: 0.792 (ATM) -> 0.798Improved
add hierarchical decomposition, at cost of more parameters and computeViEEG vs NICE · THINGS-EEGparameters / FLOPs: 2630.940 K -> 7924.488 K params; 42.7354 M -> 127.973 M FLOPsHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy