The validation loss is decreasing very slowly and the accuracy after 100 epochs is significantly lower than the train accuracy. What seems to be the problem here and how can I fix it? Do I decrease the number of nodes in the hidden layer, or use dropout?
