Deep learning has revolutionized the performance of classification, but in many settings where either massively annotating labels is a labor-intensive task, or only limited datasets are available, a deep learning model will easily overfit and generalizes poorly. Data augmentation as a problem of training a class-conditional and supervised generative adversarial network (GAN) is elaborately formulated in a deep adversarial data augmentation (DADA) technique. Simulation were carried out on CIFAR-100, CIFAR-10 and SVNH to train deep classifiers in the extremely low data regimes showing improved performance through DADA over traditional data augmentation. Deep classifiers were trained on three real-world small datasets: the Karolinska Directed Emotional Faces (KDEF) dataset for the facial expression recognition task, a Brain-Computer Interface (BCI) Competition dataset for the EEG brain signal classification task, and the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) dataset for the tumor classification task. For all of them, DADA leads to highly competitive generalization performance.
Input variables : Images or EEG Signal
Output Variables : Artificially synthesized data samples
| Statistical | : | Somers D | Accuracy | Precision and Recall | Confusion Matrix | F1 Score | Roc and Auc | Prevalence | Detection Rate | Balanced Accuracy | Cohen's Kappa | Concordance | Gini Coefficent | KS Statistic | Youden's J Index |
| Infrastructure | : | Log Bytes | Logging/User/IAMPolicy | Logging/User/VPN | CPU Utilization | Memory Usage | Error Count | Prediction Count | Prediction Latencies | Private Endpoint Prediction Latencies | Private Endpoint Response Count |
Visit Model : github.com
Additional links : arxiv.org
| Model Category | : | Public |
| Date Published | : | August, 2018 |
| Healthcare Domain | : |
Medical Technology
Provider |
| Code | : | github.com |
| Medical Imaging |
| Image Synthesis |