Gaussian mixture models with Wasserstein distance
This work addresses a subtle training problem in generative modeling for structured real-world data, offering an incremental improvement in efficiency and control.
The paper tackles the challenge of training generative models with discrete and continuous latent variables, which often underutilize the discrete component, by using Wasserstein Autoencoders to fully leverage the discrete latent without objective modifications or fine-tuning, resulting in comparable sample generation with simpler networks and enhanced control.
Generative models with both discrete and continuous latent variables are highly motivated by the structure of many real-world data sets. They present, however, subtleties in training often manifesting in the discrete latent being under leveraged. In this paper, we show that such models are more amenable to training when using the Optimal Transport framework of Wasserstein Autoencoders. We find our discrete latent variable to be fully leveraged by the model when trained, without any modifications to the objective function or significant fine tuning. Our model generates comparable samples to other approaches while using relatively simple neural networks, since the discrete latent variable carries much of the descriptive burden. Furthermore, the discrete latent provides significant control over generation.