MNIST

This project bases on the Kaggle digital recognition competition.  
Data: From the TensorFlow’s package.  
Input: 28*28=784  
Output: 10  

Five files:

1. Autoencoder

Key Words: Xaiver distribution

2. Softmax

Use SoftMax function to train the data.
Key Words: GradientDescentOptimizer
Result: 92.05%

3. NN

Use ANN to train the data.
Key Words: epoch(5000), dropout(0.8), hidden_units(300), batch(100), ReLU
Result: 97.99%

4. Double_NN

Use two hidden layers to train the data
Key Words:  hidden_units(300, 300)
Result: 98.11%

5. CNN

Use CNN to train the data
Key Words: (conv[5,5]-pool-conv[5,5]-pool-fc[1024]-fc)
Results: 99.19%