TR2026-021

Polarization-Based Data Reuploading

    •  Taguchi, M., Koike-Akino, T., Kojima, K., "Polarization-Based Data Reuploading", The IEEE Photonics Conference (IPC), DOI: 10.1109/​IPC65510.2025.11282035, November 2025.
      BibTeX TR2026-021 PDF
      • @inproceedings{Taguchi2025dec,
      • author = {Taguchi, Mari and Koike-Akino, Toshiaki and Kojima, Keisuke},
      • title = {{Polarization-Based Data Reuploading}},
      • booktitle = {The IEEE Photonics Conference (IPC)},
      • year = 2025,
      • month = dec,
      • publisher = {IEEE},
      • doi = {10.1109/IPC65510.2025.11282035},
      • issn = {2575-274X},
      • isbn = {979-8-3315-2559-0},
      • url = {https://www.merl.com/publications/TR2026-021}
      • }
  • MERL Contact:
  • Research Area:

    Electronic and Photonic Devices

Abstract:

We propose imperfection-tolerant optical machine learning of data reuploading by polarization rotation. While the conventional optical unitary transformation method is impractical due to prohibitively high requirement in its optical coupler accuracy, the proposed method eliminates the couplers at comparable performance in a classification task.