Likely Interpolants of Generative Models
This provides a tool for controlled generation and model inspection in generative AI, though it is incremental as it builds on existing interpolation concepts.
The paper tackles the lack of a principled interpolation scheme in generative models by developing a general method that computes likely transition paths without additional training, showing it traverses higher density regions than baselines across various models and datasets.
Interpolation in generative models allows for controlled generation, model inspection, and more. Unfortunately, most generative models lack a principal notion of interpolants without restrictive assumptions on either the model or data dimension. In this paper, we develop a general interpolation scheme that targets likely transition paths compatible with different metrics and probability distributions. We consider interpolants analogous to a geodesic constrained to a suitable data distribution and derive a novel algorithm for computing these curves, which requires no additional training. Theoretically, we show that our method locally can be considered as a geodesic under a suitable Riemannian metric. We quantitatively show that our interpolation scheme traverses higher density regions than baselines across a range of models and datasets.