Use Turbowarp! https://turbowarp.org Draw a number 0-9 in the black square. Make sure it is somewhat centred and legible. The MLP will try to guess the number you have drawn. This neural net was trained in python and imported to scratch. It has a 784->48->32->16->10 architecture. Press space to reset the canvas. #mnist #ai #nets #complex #drawing #number #recognition #numbers #cool #coding #MLP
I had a little help from Claude to set up the neural net, but it was fairly simple. The training batch increases with epoch and so does learning rate. The learning rate decreases with time. The dataset is MNIST but it is augmented with some rotations, translations and skews. This helps the models recognition of messier handwriting. All of this was made in python with PyTorch.