Summary:
-Code uses keras 2.2 and 2.4.
-2.2 is used in conjunction with plaidml backend for AMD GPU.
-2.4 is used with tensorflow backend for NVIDIA GPU.
-Repository contains code for single layer NN, DNN, and CNN.
-Hyperparameters were coded so they could be adjusted with ease for testing purposes.
-Datasets used are the MNIST database, CIFAR-10, and CIFAR-100. ImageNET may be done if time/resources permits.
-Single layer NN was only tested on MNIST. Both CIFARs only contain DNN and CNN.
Testing:
o Number of hidden layers
o Epochs
o Batch size
o Dropout, batch normalisation (ordering of layers too)
o Activation functions – sigmoid, tanh, relu, softmax
o Optimisers – learning rate, momentum.
o Loss functions – cross entropy.
o And more!
下载地址:
https://url92.ctfile.com/f/1850492-582197374-0c7083?p=3660 (访问密码: 3660)