ESS-Flow: Training-free guidance of flow-based models as inference in source space
This addresses the need for flexible conditional generation in domains like materials science and biology without requiring paired data or reliable gradients, though it is incremental as it builds on existing flow-based models.
The paper tackles the problem of guiding pretrained flow-based generative models for conditional generation without retraining by introducing ESS-Flow, a gradient-free method that uses Elliptical Slice Sampling in the source space, achieving effective results in materials design and protein structure prediction.
Guiding pretrained flow-based generative models for conditional generation or to produce samples with desired target properties enables solving diverse tasks without retraining on paired data. We present ESS-Flow, a gradient-free method that leverages the typically Gaussian prior of the source distribution in flow-based models to perform Bayesian inference directly in the source space using Elliptical Slice Sampling. ESS-Flow only requires forward passes through the generative model and observation process, no gradient or Jacobian computations, and is applicable even when gradients are unreliable or unavailable, such as with simulation-based observations or quantization in the generation or observation process. We demonstrate its effectiveness on designing materials with desired target properties and predicting protein structures from sparse inter-residue distance measurements.