Deep Residual Mixture Models
This addresses the need for more interactive and exploratory machine learning by reducing user waiting time for retraining, though it appears incremental as an architectural improvement.
The authors tackled the problem of limited flexibility in conditional sampling of deep generative models by proposing Deep Residual Mixture Models (DRMMs), which enable training once and then sampling with arbitrary conditioning variables, Gaussian priors, and constraints, as demonstrated in constrained multi-limb inverse kinematics and controllable animation generation.
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 one should minimize the user waiting for retraining a model. We demonstrate DRMMs in constrained multi-limb inverse kinematics and controllable generation of animations.