Scheduled denoising autoencoders

TL;DR

Scheduled denoising autoencoders progressively reduce noise during training to learn multi-scale features, significantly improving supervised task performance and achieving state-of-the-art results on CIFAR-10.

Contribution

Introduces a novel training schedule for denoising autoencoders that enhances feature learning across multiple scales and improves downstream task accuracy.

Findings

Boosts supervised task performance compared to standard autoencoders.

Achieves lowest reported error on CIFAR-10 among permutation-invariant methods.

Learns both coarse and fine features depending on noise level during training.

Abstract

We present a representation learning method that learns features at multiple different levels of scale. Working within the unsupervised framework of denoising autoencoders, we observe that when the input is heavily corrupted during training, the network tends to learn coarse-grained features, whereas when the input is only slightly corrupted, the network tends to learn fine-grained features. This motivates the scheduled denoising autoencoder, which starts with a high level of noise that lowers as training progresses. We find that the resulting representation yields a significant boost on a later supervised task compared to the original input, or to a standard denoising autoencoder trained at a single noise level. After supervised fine-tuning our best model achieves the lowest ever reported error on the CIFAR-10 data set among permutation-invariant methods.