Recent and Yet-to-Be-Explored GAN Topics

In this lesson, we will cover a few recent and yet-to-be-explored GAN topics that are challenging, interesting, and valuable.

Verified AI

One of the most interesting topics in GANs and deep learning is verified AI. This topic was described in Sanjit Seshia’s “Towards Verified AI” paper in 2016 and was later addressed in a blog post by Google’s DeepMind team. There are many challenges involved in achieving verified AI. Some of these challenges include testing, training, and formally proving that the models are specification-consistent.

GAN applications in biology

Other fields that have recently received attention from GAN researchers include biology and its related subfields. There are GAN models that address the problem of drug discovery and real-valued time series generation, as seen in the papers “3D Molecular Representations Based on the Wave Transform for Convolutional Neural Networks” and “Real-Valued (Medical) Time Series Generation with Recurrent Conditional GANs,” respectively.

Audio synthesis with GANs

Audio synthesis has also received some attention in the GAN community. GANSynth and WaveGAN are GAN models that have been used for audio synthesis. Chris Donahue’s website has an interesting demo that uses WaveGAN to fuel a drum loop.

GANs have been used in many other tasks, such as estimating individualized treatment effects (GANITE), multivariate time-series imputation (as seen in the paper “Multivariate Time Series Imputation with Generative Adversarial Networks”), autonomous driving (D-GAN, DeepRoad, and SADGAN), and more.