A Bayesian Data Augmentation Approach for Learning Deep Models

TL;DR

This paper introduces a Bayesian data augmentation method for deep learning that models new training samples as missing data and generates them based on learned distributions, improving classification performance.

Contribution

It proposes a novel Bayesian formulation for data augmentation using a generalized Monte Carlo EM algorithm and extends GANs to implement this approach.

Findings

Outperforms traditional augmentation methods on MNIST, CIFAR-10, CIFAR-100

Produces better classification results than similar GAN models

Demonstrates the effectiveness of Bayesian data augmentation in deep learning

Abstract

Data augmentation is an essential part of the training process applied to deep learning models. The motivation is that a robust training process for deep learning models depends on large annotated datasets, which are expensive to be acquired, stored and processed. Therefore a reasonable alternative is to be able to automatically generate new annotated training samples using a process known as data augmentation. The dominant data augmentation approach in the field assumes that new training samples can be obtained via random geometric or appearance transformations applied to annotated training samples, but this is a strong assumption because it is unclear if this is a reliable generative model for producing new training samples. In this paper, we provide a novel Bayesian formulation to data augmentation, where new annotated training points are treated as missing variables and generated based on the distribution learned from the training set. For learning, we introduce a theoretically sound algorithm --- generalised Monte Carlo expectation maximisation, and demonstrate one possible implementation via an extension of the Generative Adversarial Network (GAN). Classification results on MNIST, CIFAR-10 and CIFAR-100 show the better performance of our proposed method compared to the current dominant data augmentation approach mentioned above --- the results also show that our approach produces better classification results than similar GAN models.