only about 0.5 val_acc

[fiasco-zh]

Hi, I have trained your model, but it is much over-fitting, more than 0.99 train-acc, and about 0.55 val-acc. Is this your final model and hy-parameters? I have trained some model architecture before, and there always be over-fitting even with using batch-norm, regularize and dropout. Do you have any advice?