基于硬件-in-the-loop 训练的 4f 光相关器,采用对数复杂度降低用于 CNN
神经与进化计算
2025-01-09 v1
摘要
本文评估了在 MNIST 数据集上进行前向-only 学习算法的硬件-in-the-loop 训练,实现了 87.6% 的准确率,复杂度为 O(n²),与反向传播相比后者实现 88.8% 的准确率,复杂度为 O(n² log n)。
关键词
引用
@article{arxiv.2501.04141,
title = {Hardware-In-The-Loop Training of a 4f Optical Correlator with Logarithmic Complexity Reduction for CNNs},
author = {Lorenzo Pes and Maryam Dehbashizadeh Chehreghan and Rick Luiken and Sander Stuijk and Ripalta Stabile and Federico Corradi},
journal= {arXiv preprint arXiv:2501.04141},
year = {2025}
}