Module Leader: “This section won’t be significant in final exam”

Intro

Autoencoders (also known as Neural Network Autoencoders) are designed to reproduce the input, especially for images.

• It’s an unsupervised learning method
• It can learn nonlinear dependencies
• It can use convolutional layers
• It can use transfer learning

Structure

Applications

Properties

Overfitting & Solution

PCA vs. Autoencoder

Application in document compression & retrieval