Dreaming More Data: Class-dependent Distributions over Diffeomorphisms for Learned Data Augmentation

TL;DR

This paper introduces a method for learning class-specific diffeomorphic transformations for data augmentation, enabling more effective and automated augmentation strategies that improve deep neural network training.

Contribution

It proposes a novel approach to learn class-dependent transformations as probabilistic models on a Riemannian submanifold of diffeomorphisms, moving beyond manual augmentation schemes.

Findings

Significant performance improvements in neural network training.

Automated, class-specific transformation learning.

Enhanced data augmentation effectiveness.

Abstract

Data augmentation is a key element in training high-dimensional models. In this approach, one synthesizes new observations by applying pre-specified transformations to the original training data; e.g.~new images are formed by rotating old ones. Current augmentation schemes, however, rely on manual specification of the applied transformations, making data augmentation an implicit form of feature engineering. With an eye towards true end-to-end learning, we suggest learning the applied transformations on a per-class basis. Particularly, we align image pairs within each class under the assumption that the spatial transformation between images belongs to a large class of diffeomorphisms. We then learn a class-specific probabilistic generative models of the transformations in a Riemannian submanifold of the Lie group of diffeomorphisms. We demonstrate significant performance improvements in training deep neural nets over manually-specified augmentation schemes. Our code and augmented datasets are available online.