Deep Adversarial Data Augmentation For Extremely Low Data Regimes

The performance of classification and recognition has been tremendously revolu-tionized by the prosperity of deep learning.Deep learning-based classifiers and detec-tors can reach unprecedented accuracy given that there are 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 avail-able,and nothing-including unlabeled data-else.Such regimes arise from practical situations where not only data labeling but also data collection itself is expensive.Most existing methods in the low data regimes deal with the scarcity of labeled data;however,they often assume the help from abundant unlabeled samples in the same set,or(labeled or unlabeled)samples from other similar datasets,enabling various semi-supervised or transfer learning solutions.Different from them,This thesis proposes a deep adversarial data augmentation(DADA)technique to address the problem,in which this thesis elaborately formulates data augmentation as a problem of training a class-conditional and supervised genera-tive 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.To quantitatively validate its effective-ness,this thesis first performs extensive simulations to show that DADA substantially outperforms both traditional data augmentation and a few GAN-based options.This thesis then extends experiments to four real-world small labeled classification datasets where existing data augmentation and/or transfer learning strategies are either less ef-fective or infeasible.This thesis also demonstrates that DADA to can be extended to the detection task.This thesis improves the pedestrian synthesis work by substitute for our discriminator and training scheme.Validation experiment shows that DADA can improve the detection mean average precision(mAP)compared with some traditional data augmentation techniques in object detection.