From Autoencoders To Variational Autoencoders: Improving The Encoder