BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Modelling\, Estimating and Sampling Multiscale Maximum Entropy Dis
 tributions in High Dimensions
DTSTART:20260609T091500Z
DTEND:20260609T101500Z
DTSTAMP:20260614T032700Z
UID:indico-event-14485@indico.math.cnrs.fr
DESCRIPTION:Speakers: Etienne Lempereur (ENS)\n\nModelling\, estimating an
 d sampling non-Gaussian probability distributions in high dimension from l
 imited data lies at the heart of statistical physics and machine learning.
  Such distributions describe processes as varied as ocean currents\, flock
 s of birds\, or natural images. Classical approaches rely on models that a
 re typically maximum-entropy distributions with few parameters\, limited i
 n expressivity and estimated by algorithms that do not scale to high dimen
 sions. Recent machine-learning methods\, by contrast\, sample efficiently 
 from high-dimensional\, multimodal distributions by transporting noise ont
 o data—but rely on neural networks with billions of parameters\, limitin
 g interpretability and requiring hundreds of thousands of training samples
 . \nIn this presentation\, we develop algorithms to estimate and sample h
 igh-dimensional maximum-entropy distributions—non-Gaussian\, multiscale 
 and multimodal—from few realisations. The central challenge is the curse
  of dimensionality\, which afflicts modelling\, estimation and sampling al
 ike because non-Gaussian processes in physics and nature exhibit long-rang
 e dependencies and multimodal distributions. On one hand\, a hierarchical 
 factorisation of the distribution into conditional probabilities across sc
 ales in a wavelet basis disentangles long-range dependencies.  On the oth
 er hand\, multimodality can be adressed with Moment-Guided Diffusion (MGD)
 \, an algorithm that transports noise onto the target maximum-entropy dist
 ribution. Both strategies define a mathematically guaranteed framework to 
 model with few parameters\, estimate from scarse datasets and sample effic
 iently multiscale maximum entropy distributions in high dimensions\, with 
 applications to physics and finance.\n\nhttps://indico.math.cnrs.fr/event/
 14485/
LOCATION:Salle J. Cavailles (1R2-132)
URL:https://indico.math.cnrs.fr/event/14485/
END:VEVENT
END:VCALENDAR