Adversarial Training Improves Joint Energy-Based Generative Modelling
This work addresses generative modeling challenges for researchers in machine learning, but appears incremental as it builds on existing energy-based and adversarial training methods.
The authors tackled the problem of generative modeling by proposing a hybrid energy-based framework that combines robust classifier gradients with Langevin Dynamics, resulting in improved training stability, robustness, and generative performance.
We propose the novel framework for generative modelling using hybrid energy-based models. In our method we combine the interpretable input gradients of the robust classifier and Langevin Dynamics for sampling. Using the adversarial training we improve not only the training stability, but robustness and generative modelling of the joint energy-based models.