About This Event:
The Boston Machine Intelligence Reading Group is a recurring event series organized by Cambrio and General Assembly, designed to explore the latest research in machine learning and artificial intelligence. Join us to discover great new research, get together with fellow researchers/scientists/engineers, learn, and share ideas.
Improved Training of Wasserstein GANs
In August, we'll discuss 'Improved Training of Wasserstein GANs' By Ishann Gulrajani, et al.
Why It Matters?
Generative adversarial networks are a promising new kind of model that can learn from unsupervised + unstructured data like images, text, and sequences. But because GANs are notoriously difficult to train, applications have been limited to a specific highly-tuned model architecture and task. This paper reveals a way to stabilize GAN training. The method seems to work robustly, with strong results for a variety of tasks, from images (e.g., 101-layer ResNet) to text (e.g., character-level language model).