Learning to Compose Domain-Specific Transformations for Data Augmentation

By Alexander J. Ratner, Henry R. Ehrenberg, Zeshan Hussain, Jared A. Dunnmon, Christopher R'e. 2017

367 citations recorded by the source corpus.

Data augmentation is a ubiquitous technique for increasing the size of labeled training sets by leveraging task-specific data transformations that preserve class labels. While it is often easy for domain experts to specify individual transformations, constructing and tuning the more sophisticated compositions typically needed to achieve state-of-the-art results is a time-consuming manual task in practice. We propose a method for automating this process by learning a generative sequence model over user-specified transformation functions using a generative adversarial approach. Our method can make use of arbitrary, non-deterministic transformation functions, is robust to misspecified user input, and is trained on unlabeled data. The learned transformation model can then be used to perform data augmentation for any end discriminative model. In our experiments, we show the efficacy of our approach on both image and text datasets, achieving improvements of 4.0 accuracy points on CIFAR-10, 1.4 F1 points on the ACE relation extraction task, and 3.4 accuracy points when using domain-specific transformation operations on a medical imaging dataset as compared to standard heuristic augmentation approaches.

Subfields: Artificial Intelligence, Information Systems, Representation & Self-Supervised Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

In Representation & Self-Supervised Learning

Citations in the corpus

Randaugment: Practical automated data augmentation with a reduced search space (2019, methodology): “Furthermore, generative adversarial networks (GAN) have been used to choose optimal sequences of data augmentation operations[39].”
A Survey of Data Augmentation Approaches for NLP (2021, background): “Models can also learn to combine together simpler DA primitives (Cubuk et al., 2018; Ratner et al., 2017) or add”
AutoAugment: Learning Augmentation Policies from Data (2018, methodology): “An exception is work by Ratner et al., who used GANs to generate sequences that describe data augmentation strategies [50].”
Faster AutoAugment: Learning Augmentation Strategies using Backpropagation (2019, methodology): “One direction is to search for better combinations of symbolic operations using black-box optimization techniques such as reinforcement learning [5,24], evolution strategy [32], Bayesian optimization [18] and population-based training [12].”
Automatically Learning Data Augmentation Policies for Dialogue Tasks (2019, background): “…have also adopted Generative Adversarial Networks (Goodfellow et al., 2014) to either directly generate augmented data (Tran et al., 2017; Sixt et al., 2018; Antoniou et al., 2017; Zhu et al., 2017; Mun et al., 2017; Wang and Perez, 2017), or generate augmentation strategies (Ratner et al., 2017).”
Albumentations: fast and flexible image augmentations (2018, background): “Different domains, imaging modalities, and tasks may benefit from a wide range of different amounts and combinations of image transformations [4]–[6].”
Learning Data Augmentation Strategies for Object Detection (2019, methodology): “, used generative adversarial networks to generate sequences of data augmentation operations [37].”
Disentangling Adversarial Robustness and Generalization (2018, background): “, [74, 91, 4, 20], where training examples are generated, but without the goal to be misclassified.”
Bias Correction of Learned Generative Models using Likelihood-Free Importance Weighting (2019, background): “These new datapoints are derived from an existing labeled dataset, either by manual transformations (e.g., rotations, ﬂips for images), or alternatively, learned via a generative model [1, 20].”
Joint Contrastive Learning with Infinite Possibilities (2020, background): “Recent works [1, 4, 25, 38] show that semantic data augmentation is able to effectively preserve the class identity.”
The Role of Massively Multi-Task and Weak Supervision in Software 2.0 (2019, background): “For example, in our prototype Software 2.0 system Snorkel, rather than simply loading data points and labels from a static file during training, data points are loaded as complex, structured objects via a SQL query, and then weakly labeled and potentially augmented [30] during training.”
How to Train Your MAML to Excel in Few-Shot Classification (2021, methodology): “Meta-learning has been applied in various fields, including imbalance learning [74, 106], data compression [103], architecture search [26], recommendation systems [96], data augmentation [69], teaching [27], and hyper-parameter tuning [33, 66].”
On the Generalization Effects of Linear Transformations in Data Augmentation (2020, neutral): “By searching over a (possibly large) space of transformations through reinforcement learning based techniques, augmentation schemes including AutoAugment (Cubuk et al., 2018) and TANDA (Ratner et al., 2017) have shown remarkable gains over various models on image classiﬁcation tasks.”