Welcome back to this series about digits recognition! Today, we’re entering the difficult-but-rewarding part of the process. We’ll turn our randomly-generated neural network into a trained expert of digits recognition.
How do we teach a neural network?
Finding the right weights and biases
Gradient descend (the maths are back)
Computing the gradient with backpropagation
Epochs, batches, etc.
Coding it all
The structure
The main function: train
Updating weights and biases with learn
The final piece: backpropagation
Outline
Forward pass
Backward pass
Reward time
Testing on a few images
What’s left to do
