I made a neural network and trained it in scratch to recognize single drawn digits. The database contains ~700 hand-drawn digits. The training phase had a duration of 15 minutes I think. You can retrain it, everything is in the project if you want to tinker with it. It's not very accurate, but i guess it works. -Input is an 16x16 image -Hidden layer is 16 neurons -Output layer is 10 neurons (one for each digit) -True output (prediction) is the highest value of the output layer Used sigmoid as an activation function and trained with backpropagation. That's it I guess. #NeuralNetwork #AI #DeepLearning
