输入在已发布的 /100 缩放之前,先去掉每个片段各通道的均值(评分前已声明);不去均值的敏感性分析中,P300 的得分高 0.71、睡眠低 1.40(单位:百分点)。Input removes the per-segment per-channel mean before the published /100 scaling (declared before scoring); a no-mean-removal sensitivity run scores P300 0.71 percentage points higher and sleep 1.40 lower.
输入先去掉每个片段各通道的均值(评分前已声明);不去均值的敏感性分析中,P300 的得分低 2.06、睡眠低 0.93(单位:百分点)。Input removes the per-segment per-channel mean (declared before scoring); the no-mean-removal sensitivity run scores P300 2.06 percentage points lower and sleep 0.93 lower.
重采样到 256 Hz,并截断为完整的 40 采样点片段;1–2 秒的时间窗和 4–8 通道的电极布局,都超出了作者测试过的范围。Resampled to 256 Hz and truncated to whole 40-sample patches; 1-2 s windows and 4-8 channel montages are outside the authors' tested regime.
同 LUNA Base。只做冻结探针。As LUNA Base. Frozen probes only.
p300-target 与 semantic-target 未运行:分词器需要 2 秒的时间窗,1 秒的片段不补零就无法满足。在 4–8 通道的电极布局上保留了上游的共同平均参考。p300-target and semantic-target not run: the 2 s tokenizer window cannot be met by 1 s segments without zero padding. Upstream common average reference kept on 4-8 channel montages.
沿用已发布的 CBraMod 池化方式(对片段取均值,保留通道);上游的注意力窗口只有 1 个 token,所以 LoRA 实际上只适配了 V 投影。Published CBraMod pooling (patch mean, channels kept); LoRA effectively adapts only the V projection (1-token attention window upstream).
BETA 的单元格:数据出现在预训练数据中(BETA 在 ST-EEGFormer 的预训练语料里)。BETA cells are pretraining-exposed (BETA is in the ST-EEGFormer corpus).
BETA 的单元格:数据出现在预训练数据中。这里使用作者的下游类(时间索引从 1 开始);Large 预训练时用的可能是从 0 开始的索引。只做冻结探针。BETA cells are pretraining-exposed. The authors' downstream class (1-based temporal index) was used; Large may have been pretrained with a 0-based index. Frozen probes only.
论文推荐的掩码几何;保留通道的池化(C × 512)。另外三个同系列检查点列在掩码消融面板中。Paper-recommended masking geometry; channel-retained pooling (C x 512). Three sibling checkpoints appear in the masking ablation panel.
设计范围内:p300-target 与 semantic-target。其余六个协议是模型设计范围之外的阴性对照。ERP 时间窗没有刺激前基线。In design: p300-target and semantic-target. The other six protocols are negative controls outside the model's design. ERP windows have no pre-stimulus baseline.
semantic-target 的数据出现在预训练数据中(TMNRED 在 SingLEM 的预训练语料里)。每个通道单独编码;每个 2 秒片段的最后 0.25 秒,分词器看不到。semantic-target is pretraining-exposed (TMNRED is in the SingLEM corpus). Each channel is encoded independently; the last 0.25 s of each 2 s segment is not seen by the tokenizer.
每个核心数据集是否出现在预训练数据中都未知(作者没有公开来源清单)。只用编码器;不做通道上采样或重建。Pretraining exposure unknown for every core dataset (no published source list). Encoder only; no channel upsampling or reconstruction.
论文推荐检查点在掩码消融中的同系列检查点:架构、语料与训练方案相同,掩码框架或几何不同。A sibling of the paper-recommended checkpoint in the masking ablation: the same architecture, corpus and recipe, another masking framework or geometry.
REVE Base · REVE Large:REVE 负责任使用许可 v1.0(站点所有者于 2026-10-04 批准)REVE Responsible Use License v1.0 (owner-approved 2026-10-04)REVE 负责任使用许可 v1.0:只发布聚合的科学结果;不做重新识别、模型反演、成员推断或未经同意的画像分析;引用 REVE 论文并写明模型版本;不分享任何适配后的权重。REVE Responsible Use License v1.0: aggregate scientific results only; no re-identification, model inversion, membership inference or non-consensual profiling; cite the REVE paper and name the model version; no adapted weights shared.模型版本:REVE Base @ dc2a075c · REVE Large @ 317531c7。论文 ↗
LUNA Base · LUNA Large:CC BY-ND 4.0(经站点所有者批准:仅限内部适配,只发布聚合分数,不分享任何适配后的权重或 LoRA 增量)CC BY-ND 4.0 (owner-approved: internal adaptation, aggregate scores only, no adapted weights or LoRA deltas shared)CC BY-ND 4.0 权重:分数只来自内部的探针与适配;不分享任何修改后的权重、LoRA 或适配器增量;不代表 LUNA 作者的认可;也不借 LUNA 之名暗示这是官方发布。CC BY-ND 4.0 weights: scores from internal probing/adaptation only; no modified weights, LoRA or adapter deltas are shared; no implied endorsement by the LUNA authors; the LUNA name is not used to suggest an official release.论文 ↗
ST-EEGFormer Base · ST-EEGFormer Large:MIT(仓库的 LICENSE;README 保留论文、示意图与名称的权利)MIT (repo LICENSE; README reserves the paper, diagrams and the name)MIT(仓库的 LICENSE;README 保留论文、示意图与 ST-EEGFormer 名称的权利)。MIT (repository LICENSE; the README reserves the paper, diagrams and the ST-EEGFormer name).论文 ↗
eeg-fm-masking MAE (r=9 cm, L=2) · eeg-fm-masking JEPA (r=9 cm, L=2) · eeg-fm-masking MAE (r=one channel, L=1) · eeg-fm-masking JEPA (r=one channel, L=1):CC-BY-4.0(须引用 arXiv:2609.33487)CC-BY-4.0 (cite arXiv:2609.33487)CC-BY-4.0 权重:须署名(引用 arXiv:2609.33487)。CC-BY-4.0 weights: attribution required (cite arXiv:2609.33487).论文 ↗
SingLEM:MIT(检查点提交在采用 MIT 许可的仓库中)MIT (checkpoint committed in the MIT repository)MIT。论文 ↗
ZUNA 1.1:Apache-2.0;模型卡:仅供研究使用,不可用于诊断或临床Apache-2.0; model card: research use only, not for diagnosis or clinical useApache-2.0 权重;模型卡写明仅供研究使用,不可用于诊断或临床。Apache-2.0 weights; model card states research use only, not for diagnosis or clinical use.论文 ↗
任何模型的作者都没有为这些结果背书;没有保留或分享任何适配后的权重或 LoRA 增量。
这些行的说明
REVE Base:REVE 的输入去掉了每个片段各通道的均值(评分前已声明;已发布的 LaBraM/CBraMod 行没有这样做);不去均值的敏感性分析见该行脚注。REVE input removes the per-segment per-channel mean (declared before scoring; the published LaBraM/CBraMod rows did not); a no-mean-removal sensitivity run is stated in the row footnote.
REVE Large:REVE 的输入去掉了每个片段各通道的均值(评分前已声明);不去均值的敏感性分析见该行脚注。REVE input removes the per-segment per-channel mean (declared before scoring); a no-mean-removal sensitivity run is stated in the row footnote.
LUNA Base:LUNA 重采样到 256 Hz,只保留完整的 40 采样点片段:1 秒保留 240/256 个采样点,2 秒保留 480/512 个。LUNA resamples to 256 Hz and keeps whole 40-sample patches: 1 s keeps 240/256 samples, 2 s keeps 480/512.LUNA 在 6 通道电极布局上对电极位置做最小-最大缩放。LUNA min-max position scaling on a 6-channel montage.
LUNA Large:LUNA 重采样到 256 Hz,只保留完整的 40 采样点片段:1 秒保留 240/256 个采样点,2 秒保留 480/512 个。LUNA resamples to 256 Hz and keeps whole 40-sample patches: 1 s keeps 240/256 samples, 2 s keeps 480/512.LUNA 在 6 通道电极布局上对电极位置做最小-最大缩放。LUNA min-max position scaling on a 6-channel montage.
BrainOmni Base:BrainOmni 在 6 个通道上做共同平均参考,去掉的片段通道方差比例中位数为 0.45。BrainOmni common average reference on 6 channels removes a median 0.45 of segment channel variance.
ST-EEGFormer Large:ST-EEGFormer Large 使用作者的下游类(时间索引从 1 开始);一项不用标签的重建检查显示,Large 预训练时用的可能是从 0 开始的索引,所以它的时间正弦编码可能错开一个位置。ST-EEGFormer Large uses the authors' downstream class (1-based temporal index); a label-free reconstruction check suggests Large was pretrained with a 0-based index, so its temporal sinusoid may be shifted by one position.
ERP-FM Base:不在 ERP-FM 的设计范围内:阴性对照。Outside the ERP-FM design: negative control.
SingLEM:SingLEM 的分词器(128 Hz 下每个 token 128 个采样点,步长 96)看不到每个 2 秒片段的最后 0.25 秒。SingLEM tokenizer (128-sample tokens, stride 96 at 128 Hz) leaves the last 0.25 s of every 2 s segment unseen.
ZUNA 1.1:ZUNA 没有公开预训练数据来源:每个核心数据集是否出现在预训练数据中都未知。每个协议有 0–4 个电极坐标被截到 ±0.12 m 位置网格的边缘,与上游做法相同。ZUNA pretraining sources are not published: exposure unknown on every core dataset. 0-4 electrode coordinates per protocol are clamped to the edge of the +-0.12 m position grid as upstream does.
每个核心时间窗都比 EEGPT 预训练用的 4 秒时间窗短:1 秒或 2 秒,睡眠则是十五段各 2 秒的数据。无论时间窗多长,编码器对每个 0.25 秒片段的计算方式都相同;时间窗越短,可供平均的片段越少。Every core window is shorter than EEGPT’s 4-second pretraining window: 1 or 2 seconds, and sleep as fifteen 2-second segments. The encoder computes each 0.25-second patch the same way at any length; a shorter window gives fewer patches to average.
读出用的是本站固定的分类头,接在冻结的汇总 token 上,与已发布的 LaBraM、CBraMod 各行相同:在 2,048 个池化值上做岭回归(alpha 100),空闲协议上则是逐人的逻辑回归分类头。它不是作者在下游任务中用的分类头(可学习的空间滤波器加线性分类头);输入的时间窗也没有做 EEGPT 预训练时用的平均参考。The readout is the site’s fixed probe on the frozen summary tokens, as for the published LaBraM and CBraMod rows: ridge (alpha 100) on 2,048 pooled values, and on idle a per-person logistic head. It is not the authors’ downstream probe, a learnable spatial filter with a linear head, and the windows are fed without the average reference EEGPT was pretrained with.
EEGPT 是否出现在预训练数据中,同样按其作者公开的预训练数据清单判断,于 2026-10-12 核查。 它的预训练语料: 五个公开数据集,按论文表 1 所列:PhysioMI(PhysioNet 运动想象与运动执行数据)、HGD、TSU(清华大学 40 目标 SSVEP 基准数据集)、SEED 与 M3CV;58 个电极、256 Hz、毫伏、平均参考、4 秒时间窗。Five public datasets, as the paper’s Table 1 lists them: PhysioMI (PhysioNet motor imagery and execution), HGD, TSU (the Tsinghua 40-target SSVEP benchmark), SEED and M3CV; 58 electrodes, 256 Hz, millivolts, average reference, 4-second windows.
许可与条款
代码许可:Apache-2.0(仅限代码仓库中的源代码)Apache-2.0 (the repository’s source code only)
权重许可:尚未厘清:无法从检查点的托管记录中读到它自己的许可与条款,代码仓库的代码许可也不能证明它涵盖权重unresolved: the checkpoint’s own licence and terms could not be read from its hosting record, and the repository’s code licence is not evidence that it covers the weights
权重:不分发任何原始或转换后的权重;哈希只用于标识所用的检查点,不授予任何权利no original or converted weights are distributed; the hashes identify the checkpoint used and grant nothing
权利审查:依据所有者的决定发布,只发布聚合结果,与那项试点比较相同。检查点的许可无法读到;没有找到发布者禁止发布独立计算的聚合结果的规定;不分发任何原始或转换后的权重。Published by the owner’s decision, aggregate results only, as the pilot comparison is. The checkpoint’s licence could not be read; no publisher prohibition on independently computed aggregate results was found; no original or converted weights are distributed.
使用条款:LaBraM at Scale 的代码与权重为 MIT 许可(Pan-Junyu/LaBraM-at-Scale @ a0b0ce27);REVE Small 适用 REVE Responsible Use License v1.0(reve-small @ 3565bdcb),只发布聚合的科学结果。不分发任何权重。
Balanced quality-screened scalp subset, not whole-night deployment prevalence and not ear-EEG. Frozen encoders average fifteen 2-second representations; EEGNet receives 10 epochs.
协议步骤
5 participant-disjoint folds. All recordings from a person stay together. Each person contributes to the held-out predictions once.
Retain six named scalp electrodes; exclude mastoids. Drop any epoch with source per-channel missing-value flag in those six electrodes, actual nonfinite values, any channel std<0.01uV, or peak-to-peak>1000uV. Use full30s at200Hz, no further filtering/reference/amplitude transformation. Select up to30 evenly spaced eligible epochs per participant/class across available nights.
One fixed seed (20260919); no early stopping or test-based tuning. EEGNet trains for 10 epochs per fold. Frozen encoders use training-only standardized ridge heads (alpha 100).
Lightweight balanced quality-screened subset; scores do not describe natural sleep-stage prevalence or the entire73780epoch release.
Two original corrupt sessions and boundary epochs were excluded by the uploader.
200Hz data may have finite replacements despite original missing-value flags; source flags are therefore enforced instead of relying on finite checks alone.
One prediction per30s epoch. Frozen encoders pool15nonoverlapping2s segments; no sequence context across epochs.
All nights from a participant remain together; pretraining overlap unknown.
Quality thresholds fixed before any scores; source artifacts may remain.
稳定性
One fixed seed and training budget; multi-seed sensitivity pending.
是否出现在预训练数据中
Unknown unless explicitly documented; no unseen-pretraining claim.
One fixed configuration. Foundation encoders remain frozen; small networks train from scratch. These scores do not establish optimal fine-tuned performance. No individual predictions or participant-level results are distributed.
模型条款
Spectral ridge:Trained from scratch / deterministic reference; no third-party pretrained weights. Braindecode BSD-3-Clause; MNE/scikit-learn BSD where used.
LaBraM:Code/repository: MIT · Checkpoint: committed in that repository; no separate weight terms
CBraMod:Code: MIT · Weights: Apache-2.0 (official model card)
EEGNet:Trained from scratch / deterministic reference; no third-party pretrained weights. Braindecode BSD-3-Clause; MNE/scikit-learn BSD where used.
公开数据登记说明上游 CC0 明确,研究伦理证据清楚;只使用锁定版本的头皮信号衍生数据,只发布聚合结果,并说明来源。同意书覆盖的是研究本身而不是公开发布;数据的公开依据是一项 GDPR 匿名化评估(2026-09-22 修订)。Clear upstream CC0 and study ethics evidence; use only the pinned scalp-signal derivative and publish aggregates, with provenance disclosed. Consent covered the study, not publication; the public release rests on a GDPR anonymization assessment (amended 2026-09-22).