We created multiple variations to the LeNet-5 model, for example by changing the activation function, adding a dropout or extra convolution layer.