BioFormer: Rethinking Cross-Subject Generalization via Spectral Structural Alignment in Biomedical Time-Series

ICML 2026 Poster · Forty-third International Conference on Machine Learning

Authors: Guikang Du, Haoran Li, Xinyu Liu, Zhibo Zhang, Xiaoli Gong, Jin Zhang
Affiliation: College of Computer Science, Nankai University, Tianjin, China

📄 Paper: Preprint
🌐 OpenReview: ICML 2026 OpenReview

---

🔍 Idea

BioFormer studies cross-subject generalization in biomedical time-series classification by modeling spectral drift across subjects and aligning frequency structures instead of directly aligning raw signals.

---

🏗️ Model

---

🔗 Relation to Official Repository

This repository is the personal version of the BioFormer project, maintained by Guikang Du for lightweight reproduction, experimentation, and further extensions.

For the official implementation and updates maintained by our research group, please refer to:

👉 Official Repository:
https://github.com/NKU-EmbeddedSystem/BioFormer

The official repository provides:

• More complete documentation and updates
• Standardized experimental pipelines
• Long-term maintenance and support

This personal repository mainly focuses on:

• Lightweight reproduction
• Quick experimentation
• Individual development and extensions

---

📖 Citation

If you find this work useful, please consider citing:

@inproceedings{bioformer2026,
  title     = {BioFormer: Rethinking Cross-Subject Generalization via Spectral Structural Alignment in Biomedical Time-Series},
  author    = {Du, Guikang and Li, Haoran and Liu, Xinyu and Zhang, Zhibo and Gong, Xiaoli and Zhang, Jin},
  booktitle = {Forty-third International Conference on Machine Learning},
  year      = {2026}
}

---

⭐ Star this repository if you find BioFormer helpful.