Paper Overview
Field: Machine Learning Authors: Zhenyao Cui, Siyuan Kan, Siyang Li, Ziwei Wang, Dongrui Wu Published: 2026-08-19 arXiv: 2608.19134
Abstract
Accurate visual decoding can reveal how the brain represents visual information and recover perceived content from neural signals such as electroencephalography (EEG), with potential for neural communication. However, current EEG-to-image retrieval methods perform far below their within-subject counterparts for new users without labeled calibration, limiting real-world deployment.
To understand this gap, the authors analyze EEG features across subjects and find that different subjects preserve similar relationships among concepts but express them along different coordinate directions. They therefore propose SCORE (Subject Coordinate Recovery), a target label-free framework combining recovery-aware source training with coordinate alignment at deployment.
- During training: SCORE aligns source-subject EEG to a common image space and simulates unseen-subject recovery using only source segments.
- At deployment: Both encoders are frozen. SCORE selects reliable EEG-image landmarks via hubness-corrected matching and estimates an orthogonal transformation to recover the target subject's EEG coordinates—without needing source data or target labels.
Results
On 200-way retrieval over two public benchmarks, SCORE outperforms unadapted baselines for every target subject and achieves the best overall accuracy:
| Benchmark | Top-1 | Top-5 | Gain over strongest baseline | |---|---|---|---| | THINGS-EEG2 | 53.23% | 83.55% | +17.45 / +15.70 pp | | Alljoined-1.6M | 12.01% | 32.16% | +3.08 / +4.62 pp |
Without target labels or encoder updates, SCORE brings brain-based visual decoding closer to robust, practical, low-latency deployment across users.
Link: https://arxiv.org/abs/2608.19134