{"id":"arxiv/2602.08606","url":"https://arxiv.org/abs/2602.08606","tags":["math.OC","cs.LG","math.AP","math.PR"],"score":0,"title":"Constructive conditional normalizing flows","domain":"math.OC","source":"arxiv","authors":["Borjan Geshkovski","Domènec Ruiz-Balet"],"reviews":0,"subject":"math.OC","summary":"Motivated by applications in conditional sampling, given a probability measure $μ$ and a diffeomorphism $φ$, we consider the problem of simultaneously approximating $φ$ and the pushforward $φ_{\\#}μ$ by means of the flow of a continuity equation whose velocity field is a perceptron neural network with piecewise constant weights. We provide an explicit construction based on a polar-like decomposition of the Lagrange interpolant of $φ$. The latter involves a compressible component, given by the gradient of a particular convex function, which can be realized exactly, and an incompressible…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-02-09","reactions":{}}