In this work we implement a LeNet5 network over FashionMNIST dataset. We use (separately) a dropout, weight decay(L2 loss), and batch normalization layers to show their effect on the modle. Convergence graphs in each scenario are calculated for the train and test accuracies.