Why does the neural network train on each training set, but when I try to train on all sets at the same time, the error does not decrease?

Question

Machine learning

I solve the exclusive or problem. I don't use third party libraries.

Answer 1

In general, if you build a graph, you can see that at some point the network stops learning and the retraining effect occurs.
It is impossible to overcome overfitting completely, but you can make it happen later.
1) Use regularization (L1, L2)
2) Dropout (decimation)
3) If you work with images, then you can flip them in the generator, create a white pixel, etc.