Abstract

Previous work has shown that the difficulties in learning deep generative or discriminative models can be overcome by an initial unsupervised learning step that maps inputs to useful intermediate representations. We introduce and motivate a new training principle for unsupervised learning of a representation based on the idea of making the learned representations robust to partial corruption of the input pattern. This approach can be used to train autoencoders, and these denoising autoencoders can be stacked to initialize deep architectures. The algorithm can be motivated from a manifold learning and information theoretic perspective or from a generative model perspective. Comparative experiments clearly show the surprising advantage of corrupting the input of autoencoders on a pattern classification benchmark suite.

Keywords

Computer scienceArtificial intelligenceDiscriminative modelBenchmark (surveying)Generative grammarPerspective (graphical)Unsupervised learningDeep learningRepresentation (politics)SuiteFeature learningPattern recognition (psychology)Machine learningNoise reduction

Affiliated Institutions

Related Publications

Publication Info

Year
2008
Type
article
Pages
1096-1103
Citations
7123
Access
Closed

External Links

View on DOI.org

Social Impact

Altmetric
PlumX Metrics

Social media, news, blog, policy document mentions

Citation Metrics

7123
OpenAlex

Cite This

Pascal Vincent, Hugo Larochelle, Yoshua Bengio et al. (2008). Extracting and composing robust features with denoising autoencoders. , 1096-1103. https://doi.org/10.1145/1390156.1390294

Identifiers

DOI
10.1145/1390156.1390294