wencoast / SimpleCNNandResNet

SimpleCNN and ResNet performance comparison

Deep_learning_1st_project

Simple neural network and ResNet performance comparison

Running Environment

• python 3.6.7
• Tensorflow 1.12.0
• keras 2.1.6-tf

Running the Code for SNN

Path	Function
SNN/snn_fashion_mnist.py	The original structure and hyperparameters
SNN/snn_fashion_mnist_lr_optimizer.py	Compare Momentum and Adam with different learning rates
SNN/snn_fashion_mnist_dropout.py	Compare different dropout rates
SNN/snn_fashion_mnist_optimal.py	The final model with optimal settings

TODO List

1. Comparison of Deep network architectures Simple neural network and ResNet.[Done!]
2. Using Momentum optimizer and Adam optimizer with different learning rate.[Done!]
3. Using dropout.[Done!]
4. Using batch normalization.[Done!]
5. Using different activation functions including relu, tanh, leaky_relu, Sigmoid, etc. [Done!]
6. Using data augmentation.[Done!]
7. Using different optimizers such as ADAGRAD, ADADELTA, ADAM, RMSPROP, MOM. [Done!]
8. Using local response normalization.[Done!]