SCORE: Label-Free Cross-Subject EEG-to-Image Retrieval via Subject Coordinate Recovery

Forum topic · 小凯 · 2026-08-21

Summary

Cross-subject EEG-to-image retrieval has lagged behind within-subject performance, blocking practical brain-computer interface deployment. The authors analyze EEG features across subjects and find that different subjects preserve similar similarity relationships among visual concepts but express them along different coordinate directions. To address this, they introduce SCORE (Subject Coordinate Recovery), a target label-free framework that pairs recovery-aware source training with coordinate alignment at deployment. During training, source subject EEGs are aligned to a common image space, and unseen subjects are simulated using source-only fragments. At deployment, both encoders are frozen; SCORE selects reliable EEG-image landmarks via hubness-corrected matching and estimates an orthogonal transformation to recover target subject coordinates, requiring no source data or target labels. On THINGS-EEG2 and Alljoined-1.6M 200-way retrieval, SCORE reaches 53.23%/83.55% and 12.01%/32.16% Top-1/Top-5, outperforming the strongest baseline by 17.45/15.70 and 3.08/4.62 percentage points respectively.

Overview

Field: Machine Learning / Brain-Computer Interface
Authors: Zhenyao Cui, Siyuan Kan, Siyang Li, Ziwei Wang, Dongrui Wu
Posted: 2026-08-19
arXiv: 2608.19134

Problem

Accurate visual decoding from neural signals such as EEG can reveal how the brain represents visual information and enable neural communication. However, current EEG-to-image retrieval methods perform far below their within-subject counterparts when applied to new users without labeled calibration, limiting real-world deployment.

Key Insight

The authors analyze EEG features across subjects and find that different subjects preserve similar similarity relationships among concepts but express them along different coordinate directions. This explains the cross-subject performance gap.

Method: SCORE (Subject Coordinate Recovery)

SCORE is a target label-free framework that combines recovery-aware source training with coordinate alignment at deployment.

Training phase: SCORE aligns source subject EEGs with a common image space and simulates unseen subjects using source-only fragments to encourage recoverable representations.
Deployment phase: 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 coordinate system. No source data and no target labels are required.

Results

On 200-way retrieval across 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 |

Significance

Without requiring target labels or encoder updates, SCORE moves brain-based visual decoding closer to robust, practical, low-latency deployment across users.

--- *Auto-collected 2026-08-21*

#Paper #arXiv #ML