Digit Recognition using 2 layer convolutional Neural Network on Keras - Tensorflow
Resulting accuracy on 5th epoch: Training Accuracy: 98.20% Test accuracy: 99.04% Train on 60000 samples, validation on 10000 samples
Parameter tuning needed. filter sizes need to change, code can be optimized for computation. Soon I will upload a new code, with better amalgam of networks and parameters.
conv2d_1 (Conv2D) (None, 22, 22, 100) 5000
conv2d_2 (Conv2D) (None, 19, 19, 10) 16010
max_pooling2d_1 (MaxPooling2 (None, 9, 9, 10) 0
dropout_1 (Dropout) (None, 9, 9, 10) 0
flatten_1 (Flatten) (None, 810) 0
dense_1 (Dense) (None, 128) 103808
dropout_2 (Dropout) (None, 128) 0
dense_2 (Dense) (None, 10) 1290
Total params: 126,108 Trainable params: 126,108 Non-trainable params: 0