NEAR: Anchoring Neural and Visual Representations for Few-Trial Brain-to-Image Retrieval

Forum topic · 小凯 · 2026-08-21

Summary

A new paper (arXiv 2608.19128) by Zhenyao Cui, Siyuan Kan, Dingkun Liu, and Dongrui Wu introduces NEAR, a neural-anchor-based retrieval framework for brain-to-image decoding. Conventional brain-to-image retrieval averages many (up to 80) neural trials per image to suppress noise, which increases latency, cost, and user burden; accuracy drops sharply when only one or a few repetitions are available. The authors identify a non-transitive alignment pattern: low-repetition query signals and image representations each align with the high-repetition center but not directly with each other, indicating that query noise is only part of the problem and that gallery placement also matters. NEAR treats the high-repetition center as an anchor and approaches it from both sides: a denoiser pulls noisy queries toward the true anchor, while a small network predicts a pseudo-anchor for each candidate and pulls its image toward it. Across four datasets spanning EEG, MEG, and fMRI, NEAR consistently improves retrieval in few-repetition regimes; on THINGS-EEG2 it boosts 200-way Top-1 accuracy by 5.7 and 9.3 percentage points when averaging 1 and 4 repetitions respectively, reducing reliance on repeated stimulus presentation.

Paper Overview

Field: Machine Learning
Authors: Zhenyao Cui, Siyuan Kan, Dingkun Liu, Dongrui Wu
Published: 2026-08-19
arXiv: 2608.19128

Abstract

Decoding visual information from brain signals probes neural representations and enables neuro-rehabilitation and dream decoding. Recent brain-to-image retrieval approaches have achieved promising performance, typically by averaging many (up to 80) neural trials per image, requiring repeated stimulus presentation that increases latency, cost, and user burden. When only one or a few repetitions are available, the retrieval accuracy drops sharply.

This drop is commonly attributed to query noise, since averaging suppresses noise and increases signal stability. However, the authors find a non-transitive alignment pattern: the low-repetition query signal and the image representation each align with the high-repetition center, but not directly with each other. This shows that query noise is only part of the problem — gallery placement also affects retrieval.

The NEAR Framework

Based on this insight, the authors propose NEAR (NEural Anchor-based Retrieval), which treats the high-repetition center as an anchor and approaches it from both sides:

A denoiser pulls noisy queries toward the true anchor.
A small network predicts a pseudo-anchor for each candidate image and pulls the image representation toward it.

Results

Across four datasets spanning EEG, MEG, and fMRI, NEAR consistently improves retrieval in few-repetition regimes. On THINGS-EEG2, when averaging 1 and 4 repetitions, NEAR improves 200-way Top-1 accuracy by 5.7 and 9.3 percentage points, respectively.

By anchoring neural and visual representations, NEAR reduces reliance on repeated stimulus acquisition and moves neural retrieval closer to real-world deployment.

---

*Auto-collected on 2026-08-21*