This paper describes the transformation of a traditional in-silico classification network into an optical fully convolutional neural network with high-resolution feature maps and kernels. When using the free-space 4f system to accelerate the inference speed of neural networks, higher resolutions of feature maps and kernels can be used without the loss in frame rate. We present FatNet for the classification of images, which is more compatible with free-space acceleration than standard convolutional classifiers. It neglects the standard combination of convolutional feature extraction and classifier dense layers by performing both in one fully convolutional network. This approach takes full advantage of the parallelism in the 4f free-space system and performs fewer conversions between electronics and optics by reducing the number of channels and increasing the resolution, making the network faster in optics than off-the-shelf networks. To demonstrate the capabilities of FatNet, it trained with the CIFAR100 dataset on GPU and the simulator of the 4f system, then compared the results against ResNet-18. The results show 8.2 times fewer convolution operations at the cost of only 6% lower accuracy compared to the original network. These are promising results for the approach of training deep learning with high-resolution kernels in the direction towards the upcoming optics era.
翻译:本文描述传统的硅内部分类网络转换成光学完全进化神经网络,具有高分辨率特征地图和内核。当使用自由空间4f系统加速神经网络的推断速度时,可以使用高分辨率的地貌地图和内核的分辨率,而不会损失框架速率。我们提供FatNet用于图像分类,这比标准革命级分类器更适合自由空间加速。它忽视了螺旋特征提取和分类器密集层的标准组合,同时在一个完全进化的网络中运行。这一方法充分利用了4f自由空间系统中的平行主义,并通过减少频道数量和增加分辨率来减少电子和光学之间的转换,使光学网络比现成网络更快。为了展示FatNet的能力,我们用CIFAR100数据集对GPU和4f系统模拟器进行了培训,然后对比了ResNet-18的结果。这一方法充分利用了4f自由空间系统中的平行系统,从而减少了电子和光学之间的平行转换,通过减少电子和光学方法的8.2倍,而原始的深度学习结果仅低于分辨率网络。这些深度学习结果的8.2 %。