[TuG3] Algorithms for Photonic Computing

2:15 PM - 2:30 PM JST(5:15 AM - 5:30 AM UTC)

[TuG3-2] An MZI-based Block Optical Neural Network based on LDU Structure

○Qi Chen¹, Dewei wang¹, Zili Cai¹, Tian Zhang¹, Jian Dai¹, Kun Xu¹ (1.State Key Laboratory of Information Photonics and Optical Communications, Beijing University of Posts and Telecommunications)

Keywords:

Optical neural networks and reservoir computing,Programmable,reconfigurable photonics

We propose an LDU-based block optical neural network, significantly reducing hardware complexity and improving matrix mapping fidelity. It achieves 90.47% accuracy on SVHN dataset and maintains 80.84% under hardware errors, demonstrating effectiveness and exceptional robustness.