Bin Shi

Error analysis of a 3-Layer SOA-based photonic deep neural network for image classification

2019 · TU/e Research Portal

We demonstrate Iris flower classification with a trained 3-layer photonic deep neural network, which is implemented by a monolithically integrated SOA-based InP cross-connect chip, achieving a final accuracy of 85.8%. An in-depth analysis of the …

Large-Scale Photonic Integrated Cross-Connects for Optical Communication and Computation

2020

An 8×8 InP cross-connect chip for optical switching within ROADMs is employed for demonstrating optical feed-forward neural networks for analog data processing. An all-optical approach is also explored for deeper optical neuromorphic computing on chip.

On Learning Rates and Schrödinger Operators

2020 · arXiv (Cornell University)

The learning rate is perhaps the single most important parameter in the training of neural networks and, more broadly, in stochastic (nonconvex) optimization. Accordingly, there are numerous effective, but poorly understood, techniques for tuning the …

Revisiting the acceleration phenomenon via high-resolution differential equations

2022 · arXiv (Cornell University)

Nesterov's accelerated gradient descent (NAG) is one of the milestones in the history of first-order algorithms. It was not successfully uncovered until the high-resolution differential equation framework was proposed in [Shi et al., 2022] that …