Fig. 5: ONN training and results. | Nature Communications

Fig. 5: ONN training and results.

From: Graphene/silicon heterojunction for reconfigurable phase-relevant activation function in coherent optical neural networks

Fig. 5

a Data examples of MNIST dataset. b Training loss and validation accuracy results using different activation functions on the MNIST dataset. c Confusion matrix using optical activation function 3 on the MNIST dataset. d Visualized activation map trained on the MNIST dataset with the optical activation function 3. e Data examples of CIFAR-10 dataset. f Training loss and validation accuracy results using different activation functions on the CIFAR-10 dataset. g Confusion matrix using optical activation function 3 on the CIFAR-10 dataset. h Visualized activation map trained on the CIFAR-10 dataset with the optical activation function 3.

Back to article page