Synopsis

This talk motivates adversarial learning in ML from a MR researcher perspective. First, I’ll be briefly discussing some limitations of supervised learning. Next, I’ll be introducing a form of adversarial learning, generative adversarial networks – or GANs for short. Then, I’ll show how we can combine GANs with compressed sensing for the purpose of MRI reconstruction. Next, I’ll be showing some work on a fully unsupervised reconstruction method using GANs. Finally, I’ll discuss some practical considerations for those interested in training their own GAN.