Reducing the Dimensionality of Data with Neural Networks
At a glance
Showed how deep neural networks can compress complex data into compact representations while preserving useful structure.
In one paragraph
Hinton and Salakhutdinov showed how to train deep autoencoders: networks that compress data and then reconstruct it. By first training one layer at a time, they learned compact representations that outperformed a standard statistical method on tested image and document datasets.
Authors 2
Organizations
Credited for this work at publication.