Learning Dynamics of Linear Denoising Autoencoders

Abstract

Denoising autoencoders (DAEs) have proven useful for unsupervised representation learning, but a thorough theoretical understanding is still lacking of how the input noise influences learning. Here we develop theory for how noise influences learning in DAEs. By focusing on linear DAEs, we are able to derive analytic expressions that exactly describe their learning dynamics. We verify our theoretical predictions with simulations as well as experiments on MNIST and CIFAR-10. The theory illustrates how, when tuned correctly, noise allows DAEs to ignore low variance directions in the inputs while learning to reconstruct them. Furthermore, in a comparison of the learning dynamics of DAEs to standard regularised autoencoders, we show that noise has a similar regularisation effect to weight decay, but with faster training dynamics. We also show that our theoretical predictions approximate learning dynamics on real-world data and qualitatively match observed dynamics in nonlinear DAEs.

arXiv comment14 pages, 7 figures, accepted at the 35th International Conference on Machine Learning (ICML) 2018

References in corpus (9)

• Exact solutions to the nonlinear dynamics of learning in deep linear neural networks
• An Autoencoder Approach to Learning Bilingual Word Representations
• Dropout Training as Adaptive Regularization
• Generalized Denoising Auto-Encoders as Generative Models
• The loss surface of deep and wide neural networks
• Geometry of Optimization and Implicit Regularization in Deep Learning
• Analyzing noise in autoencoders and deep networks
• Exponentially vanishing sub-optimal local minima in multilayer neural networks
• Local minima in training of neural networks