Reproduce “LightRidge: An End-to-end Agile Design Framework for Diffractive Optical Neural Networks” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yingjie Li, Ruiyang Chen, Minhan Lou, Berardi Sensale-Rodriguez, Weilu Gao, Cunxi Yu
LightRidge is a framework for diffractive optical neural networks (DONNs) that automates their design, exploration, and deployment, reducing design cycles from months to days and demonstrating up to 12x emulation speedup over existing tools. It features precise, differentiable physics kernels and hardware-aware training, enabling the first all-optical image segmentation and achieving 995 fps/Watt power efficiency.
Key insights
- Precise, differentiable optical physics kernels significantly bridge the gap between numerical DONN models and physical hardware behavior, validated by experimental agreement.
- Optimized tensor-level operations for light diffraction (FFT2, iFFT2, Complex MM) are the primary drivers for substantial runtime efficiency gains in DONN emulation and training.
- Deeper DONN architectures inherently demonstrate higher robustness to physical noise (e.g., detector intensity noise) compared to single-layer networks, maintaining accuracy even with up to 5% noise levels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.