Deep Residual Mixture Models

We propose Deep Residual Mixture Models (DRMMs), a novel deep generative model architecture. Compared to other deep models, DRMMs allow more flexible conditional sampling: The model can be trained once with all variables, and then used for sampling with arbitrary combinations of conditioning variables, Gaussian priors, and (in)equality constraints. This provides new opportunities for interactive and exploratory machine learning, where the user does not have to wait for retraining a model. We demonstrate these benefits in constrained multi-limb inverse kinematics, movement planning, and image completion.