[ME1] Training for Photonic Neural Networks

12:00 PM - 12:15 PM JST(3:00 AM - 3:15 AM UTC)

[ME1-4] Differentiable Neural Architecture Search for Photonic-Aware Neural Networks

Giovanni Paolini², ○Emilio Paolini¹, Nicola Andriolli², Marco Cococcioni² (1. Scuola Superiore Sant'Anna (Italy), 2. Universitá di Pisa (Italy))

Keywords:

Photonics for neuromorphic computing and deep learning,Optical neural networks and reservoir computing

We propose a differentiable Neural Architecture Search framework for photonic-aware neural networks integrating photonic constraints and quantization-aware training. The resulting architectures outperform handcrafted models achieving 79% accuracy at 6-bit quantization, closely matching full-precision baselines.