arXiv cs.AI· Mingyu Jeon, Suwan Cho, Jae Young Suh·· 12 小时前AI 评分31
EEG 基础模型受控评估：随机初始化、频带功率与验证集选层探针
Controlled Evaluation of EEG Foundation Models: Random Initialisation, Band-Power and Validation-Selected Probes
AI 导读
一项受控评估在统一协议下测试了五个公开 EEG 基础模型和一个自研 JEPA 风格模型，覆盖五个基准与四个新增任务族。按测试分数选层会把逐层峰值平均抬高 +0.034，并在 36 个格子中的 9 个反转峰值与输出差距；改用验证集选层后，随机初始化在睡眠和情绪任务上追平或超过半数模型的预训练效果，且没有模型能在情绪或运动想象任务上胜过五频带对数功率基线。
正文 · 原文
Abstract:EEG foundation models report downstream gains, yet their protocols seldom include three controls. These are the same encoder untrained, spectral features fitted on the same data, and a layer picked without looking at test scores. We therefore evaluate five publicly released EEG foundation models and an in-house JEPA-style model under one protocol with all three controls and several seeds, on five benchmarks and four added task families. The controls change the reading in three ways. First, picking the probed layer by its test score raises the layer-wise peak by an average of +0.034 and reverses the peak-versus-output gap in 9 of 36 cells. Second, with the layer picked on validation, random initialisation matches or beats pretraining for half of the models on sleep and emotion, and no model beats a five-band log-power baseline on emotion or motor imagery. Third, after full fine-tuning only 5 of 30 pairs gain beyond twice their seed spread, and the one clear loss of -0.290 AUROC persists under matched learning rates. From these results we draw four reporting practices, and we release every cell for re-analysis.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.32067 [cs.LG] |
| (or arXiv:2610.32067v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32067 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mingyu Jeon [view email]
[v1]
Fri, 23 Oct 2026 02:11:38 UTC (623 KB)
来源：arXiv cs.AI · arxiv.org