Control Consistency Losses for Diffusion Bridges
By: Samuel Howard, Nikolas Nüsken, Jakiw Pidstrigach
Potential Business Impact:
Learns how things change from start to finish.
Simulating the conditioned dynamics of diffusion processes, given their initial and terminal states, is an important but challenging problem in the sciences. The difficulty is particularly pronounced for rare events, for which the unconditioned dynamics rarely reach the terminal state. In this work, we leverage a self-consistency property of the conditioned dynamics to learn the diffusion bridge in an iterative online manner, and demonstrate promising empirical results in a range of settings.
Similar Papers
Theoretical Closed-loop Stability Bounds for Dynamical System Coupled with Diffusion Policies
Robotics
Lets robots make faster, smarter decisions.
Predicting partially observable dynamical systems via diffusion models with a multiscale inference scheme
Machine Learning (CS)
Predicts sun's future from limited views.
Foundations of Diffusion Models in General State Spaces: A Self-Contained Introduction
Machine Learning (Stat)
Unifies how computers learn from pictures and words.