Stacked Auto-Encoder (Neural Networks and Deep Learning Project)

In this porject I took three different approaches tofeature extraction being utilized in supervised and unsupervisedlearning technique.

• The first goal of this project is presented as ain-depth comparison of stacked auto-encoder with support vectormachine and multi-layer perceptron.
• The second goal of thisproject is to implement quadratic mutual information loss whichallows to make an efficient non-parametric implementation. thecomputational complexity for each of the methods above is alsopresented in addition to the role of hyper-parameters.
• The thirdpart of the project involved implementing a classifier when thenumber of outputs from the bottle neck layer were not equal tothe true target labels.
  At the end we do an extensive review of all the three methodologies used in this project.