Control of neural transport for normalizing flows

2023/07/15 by Ruiz-Balet, Domènec, Zuazua, Enrique · 3 citations
#FOS: Mathematics #Optimization and Control (math.OC)

paper · doi:10.48550/arxiv.2307.07817

Abstract

Inspired by normalizing flows, we analyze the bilinear control of neural transport equations by means of time-dependent velocity fields restricted to fulfill, at any time instance, a simple neural network ansatz. The L¹ approximate controllability property is proved, showing that any probability density can be driven arbitrarily close to any other one in any time horizon. The control vector fields are built explicitly and inductively and this provides quantitative estimates on their complexity and amplitude. This also leads to statistical error bounds when only random samples of the target probability density are available.

Cited by

• Incremental Generation is Necessary and Sufficient for Universality in Flow-Based Modelling
• Machine Learning and Control: Foundations, Advances, and Perspectives
• Orbits and attainable Hamiltonian diffeomorphisms of mechanical Liouville equations

Related

• Normalizing flows as approximations of optimal transport maps via linear-control neural ODEs
• Control, Optimal Transport and Neural Differential Equations in Supervised Learning
• Neural ODE Control for Trajectory Approximation of Continuity Equation
• TO-FLOW: Efficient Continuous Normalizing Flows with Temporal Optimization adjoint with Moving Speed
• Neural ODE Control for Classification, Approximation, and Transport