DADA: Deep Adversarial Data Augmentation for Extremely Low Data Regime Classification

By Xiaofeng Zhang, Zhangyang Wang, Dong Liu, Qing Ling. 2018

93 citations recorded by the source corpus.

Deep learning has revolutionized the performance of classification, but meanwhile demands sufficient labeled data for training. Given insufficient data, while many techniques have been developed to help combat overfitting, the challenge remains if one tries to train deep networks, especially in the ill-posed extremely low data regimes: only a small set of labeled data are available, and nothing – including unlabeled data – else. Such regimes arise from practical situations where not only data labeling but also data collection itself is expensive. We propose a deep adversarial data augmentation (DADA) technique to address the problem, in which we elaborately formulate data augmentation as a problem of training a class-conditional and supervised generative adversarial network (GAN). Specifically, a new discriminator loss is proposed to fit the goal of data augmentation, through which both real and augmented samples are enforced to contribute to and be consistent in finding the decision boundaries. Tailored training techniques are developed accordingly. Source code is available at https://github.com/SchafferZhang/DADA

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

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

1. Zhangyang Wang
2. Shiyu Chang
3. Chongxuan Li
4. Jun Zhu
5. Kun Xu
6. Xinyu Gong
7. Joachim Denzler
8. Björn Barz

In Computer Vision and Pattern Recognition

1. Zhangyang Wang
2. Shiyu Chang
3. Jun Zhu
4. Chongxuan Li
5. Xinyu Gong
6. Björn Barz
7. Joachim Denzler
8. Kun Xu

Citations in the corpus

• AutoGAN-Distiller: Searching to Compress Generative Adversarial Networks (2020, neutral): “Our future work include evaluating AGD in more GAN-based applications such as data augmentation (Zhang et al., 2019).”
• AutoGAN: Neural Architecture Search for Generative Adversarial Networks (2019, background): “One of their most notable successes lies in generating realistic natural images with various convolutional architectures [47, 5, 61, 26, 13, 67].”
• Deep Learning on Small Datasets without Pre-Training using Cosine Loss (2019, methodology): “This problem has recently also been tackled by incorporating a GAN for data augmentation into the learning process [53].”
• Data Augmentation for Deep-Learning-Based Electroencephalography. (2020, background): “[63] GA N MI 1 400trials 3 2 NA I spectrogra m CNN Unspecif ied Unspecifi ed NA softmax 0.”
• Triple Generative Adversarial Networks (2019, methodology): “Using a Similar 13-layer CNN Classifier Datasets SVHN CIFAR10 Tiny ImageNet Number of labels 1,000 4,000 2,000 Improved-GAN [26] 25.47y DADA [28] 24.17y 33.57 ( 0:46) 29.73 ( 2:02) CNN (our code base) 22.72 ( 2:39) 46.10 ( 3:18) 24.92 ( 1:12) Triple-GAN-V1 (ours) 21.44 ( 1:95) 32.73 ( 0:86) 20.67 ( ”