Enhancing Cancer Detection with StyleGAN-2 ADA

Data augmentation for data-deficient deep neural networks.

By: Ian Stebbins, Benjamin Goldfried, Ben Maizes

Intro

Often for many domain-specific problems, a lack of data can hinder the effectiveness and even disallow the use of deep neural networks. Recent architectures of Generative Adversarial Networks (GANs), however, allow us to synthetically augment data, by creating new samples that capture intricate details, textures, and variations in the data distribution. This synthetic data can act as additional training input for deep neural …

convolutional-neural-net data-augmentation generative-adversarial medical imaging synthetic data