Stacked Denoising Autoencoders: Learning Useful Representations in a Deep Network with a Local Denoising Criterion

By P. Vincent et al

Published on Dec. 10, 2024

Read the original document by opening this link in a new tab.

Table of Contents

Abstract
1. Introduction
1.1 Notation
1.2 General setup
2. What Makes a Good Representation? From Mutual Information to Autoencoders
2.1 Retaining Information about the Input
2.2 Traditional Autoencoders (AE)
2.3 Merely Retaining Information is Not Enough
3. Using a Denoising Criterion
3.1 The Denoising Autoencoder Algorithm

Summary

The document discusses the use of Stacked Denoising Autoencoders to learn useful representations in a deep network. It explores the concept of retaining information and the limitations of traditional autoencoders. The denoising criterion is proposed as a method to extract more useful features by training the autoencoder to reconstruct clean input from corrupted versions. This approach aims to build a more robust and beneficial higher-level representation for improved classification tasks.