the ANN is ready.
https://www.youtube.com/watch?v=WxMTm0kl-Hc
loss of the ANN is 0.005 ish, and only for qwer this 4 letters, it cost too much time to train for the whole alphabet.
loss of ANN is 0.005 ish, for all letters.
During the forward pass, the filter extracts features from the input. The backward pass computes the gradient of the loss with respect to the filter values. The filter values are updated using gradient descent, making the filters adapt to the patterns in the training data.