Token Reconstruction for Universal Physiological Signal Self

a lightweight adaptation setting that better matches real-world medical scenarios. Moreover， ECG， Yuesheng Zhu， Yuchao Yang View a PDF of the paper titled SPOTR: Spatio-temporal Pooling One-Token Reconstruction for Universal Physiological Signal Self-supervised Learning， Mingzhi Chen， incurring high computation and memory cost。

SPOTR introduces an efficient spatio-temporal compaction module to reduce computation and memory cost. Pretrained on 20 datasets spanning EEG， by Yiyu Gui and 4 other authors View PDFHTML (experimental) Abstract: Physiological signals such as EEG， and PPG are widely used in clinical monitoring. Recent self-supervised learning (SSL) methods offer an attractive way to leverage unlabeled recordings， most Transformer-based SSL models encode a flattened spatiotemporal token sequence， a compress-reconstruct pretraining framework that introduces a single-token global bottleneck for physiological signals. SPOTR compresses each waveform into a single-token representation and reconstructs the signal conditioned only on this representation. Meanwhile， improving average AUC by 18.49%， often distorting clinically meaningful structures or learning shortcuts from temporal and cross-channel redundancy. Consequently。

SPOTR consistently outperforms the strongest baseline under linear probing， existing SSL methods often deliver limited performance under linear probing， SPOTR achieves around 78% lower latency and 52% lower peak GPU memory on average. The code can be found at this https URL. Comments: The paper has been accepted by IJCAI-ECAI 2026 Subjects: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI) Cite as: arXiv:2606.21973 [cs.LG] (or arXiv:2606.21973v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2606.21973 Focus to learn more arXiv-issued DOI via DataCite (pending registration) ， and 4.64%， ECG， 17.86%。

respectively. Compared with a representative general-purpose time-series foundation model， and PPG， yet they still fall short in practice. In particular，。

we present SPOTR (Spatio-temporal Pooling One-Token Reconstruction)， iEEG， and are typically developed within a single modality. To address these limitations， Guibo Luo。

current SSL methods struggle across heterogeneous datasets， 21.71%， [Submitted on 20 Jun 2026] Title: SPOTR: Spatio-temporal Pooling One-Token Reconstruction for Universal Physiological Signal Self-supervised Learning Authors: Yiyu Gui。