Fig. 6: Handwritten digit recognition task.
From: Digital-analog hybrid matrix multiplication processor for optical neural networks
a, Layer structure of the CNN to perform the HWDR task using the modified National Institute of Standards and Technology (MNIST) database. The convolution layer of the CNN is implemented using the HOP, and the rest of the network is performed offline by a desktop computer. b, Confusion matrices for the prediction results, calculated by a desktop computer and the HOP, respectively.
