Generalized Training for Neural Network Learnability: a Spectral Methods Approach

2023/04/24 by Altai Perry, Perry, Altai, Luat T. Vuong +1
Computer Science · Engineering · Physics and Astronomy · #Digital Holography and Microscopy #FOS: Physical sciences #Neural Networks and Reservoir Computing #Optical Polarization and Ellipsometry #Optics (physics.optics)

paper · pdf · doi:10.48550/arxiv.2304.12172

openalex publication_date 2023/04/24 · openalex created_date 2023/04/27 · openalex updated_date 2026/07/28

Abstract

Hybrid optical neural networks (HONNs) offload some electronic computation to optical preprocessors to achieve low-power and fast training and inference phases in machine learning tasks. Our contribution to the development of HONNs is a spectral-methods paradigm for building synthetic training data for machine-learned models. Here, our synthetic training image data does not resemble the image test data. As a result, the neural network focuses on learning specific features parameterized by the synthetic training data. Within this paradigm, a dataset's singular value decomposition entropy indicates \it learnability, i.e., how rapidly a model converges. Subsequently, we train a neural network model to rapidly learn specific features for further downstream analyses.

Related

• Hybrid training of optical neural networks
• Training Large-Scale Optical Neural Networks with Two-Pass Forward Propagation
• Spectral Neural Networks: Approximation Theory and Optimization Landscape
• A Neural Network Classifier for Spectral Pattern Recognition. On-Line versus Off-Line Backpropagation Training
• Dynamic Spectral Backpropagation for Efficient Neural Network Training