Skip to contentSkip to filters

We use cookies for analytics.

Task-Informed and Controllable Generative Flows for Inverse Problems

Santiago Segarra

Rice University

Abstract

How can a generative flow use observations and structural constraints when solving an inverse problem? Segarra describes observation-conditioned initialization for graph reconstruction, adaptive multipliers that enforce graph constraints during sampling, and a denoiser that learns finite-time flow maps. A lookahead parameter changes the balance between a conditional-mean estimate and a perceptually plausible sample.

Topics