Glossary term
Glossary term
Evaluation and Benchmarks
One of the loss functions commonly used in generative adversarial networks, based on the earth mover's distance between the distribution of generated data and real data.
Created for this library
A research team uses Wasserstein loss in its GAN variant to stabilize training compared to the original minimax loss.
A research lab uses Wasserstein loss to train generative models when standard losses produce unstable training.
An ML researcher uses Wasserstein loss as the default in GAN experiments because it correlates better with sample quality.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License