# [2019.06.25] More on Semigroups and Neural Networks
Today I tried to solve another problem related to semigroups with
deep learning. Last time I predicted unfilled positions in a
multiplication table given the known ones. Now I started to estimate
probabilities of items in different empty cells, not their exact
values. That made the task more relaxed and gave me the ability to
use more diverse network architectures. The main hindrance became
overfitting which I fought primarily by dropout, although not
sufficiently successfully.
