[ME1] Training for Photonic Neural Networks

11:30 AM - 11:45 AM JST(2:30 AM - 2:45 AM UTC)

[ME1-2] ALoRA: Hardware-Aware Fine Tuning for Photonic Large Neural Networks

○Taichi Taniguchi¹, Mitsumasa Nakajima², Kohei Ikeda³, Toshikazu Hashimoto², Satoshi Kawakami¹ (1. Kyushu University (Japan), 2. NTT Device Technology Laboratories (Japan), 3. NTT Basic Research Laboratories (Japan))

Keywords:

Highly parallelized scalable photonic computing architectures and devices,Deep learning for photonic device and applications,Optical transformer

We investigated the impact of analog errors on large-scale photonic neural network using an experimentally obtained error model. Performance degradation due to analog error can be recovered by proposed hardware-aware fine-tuning with ~1.6 % trainable parameters.