Abstract:Cross-subject generalization in biomedical time-series refers to training on data from some subjects and testing on unseen this http URL key challenge is to suppress subject specific variability in BTS this http URL existing methods implicitly suppress the variability through model building or subject adversarial learning, but rarely model it this http URL introduce spectral drift as a new perspective to characterize subject specific this http URL, BTS signals under the same label often share consistent oscillatory structure, yet exhibit subject-dependent magnitude or phase shifts in specific frequency components, which we interpret as subject-specific variability. Building on this insight, we propose this http URL its core is a Frequency-Band Alignment Module(FBAM) that generates band-wise modulation factors from the spectral distribution and adaptively adjusts amplitude and phase to align spectral structure, thereby mitigating this http URL further pair FBAM with Sample Conditional Layer Normalization, which infers normalization parameters from intrinsic signal statistics rather than subject identity, stabilizing cross-subject this http URL experiments on six datasets demonstrate that BioFormer outperforms 12 baselines, yielding absolute F1-score improvements of 6%.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.22468 [cs.LG] |
| (or arXiv:2605.22468v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.22468 arXiv-issued DOI via DataCite (pending registration) |
From: Guikang Du [view email]
[v1]
Thu, 21 May 2026 13:27:40 UTC (9,570 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。