The objective was to creat a simple neural network to recognise handwritten digits. The inbuilt digit dataset in TensorFlow was used to train and test the model. I tested different activations functions and choose SoftMax and ReLu. After initialising the network with one layer of 100 neutrons, I trained my model on ten batches of 256 images each. Final model accuracy was 97%. TensorFlow and Keras were used to create the model
bluesky314 / digit-recognition Goto Github PK
View Code? Open in Web Editor NEWNeural Network Project