BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Opening up the black box of score estimation - Sitan Chen (Harvard
  University)
DTSTART:20240719T100000Z
DTEND:20240719T110000Z
UID:TALK219070@talks.cam.ac.uk
DESCRIPTION:In recent years there has been significant interest in the the
 oretical foundations of diffusion generative modeling. One representative 
 result in this line of work is that with an accurate estimate of the score
  function for the data distribution\, one can approximately sample from vi
 rtually any bounded distribution in polynomial time. In this talk I will d
 escribe recent work on the missing piece left open by these works: when ca
 n we actually learn an accurate estimate of the score from data? I will fo
 cus on two vignettes: (1) learning Gaussian mixture models (GMMs)\, and (2
 ) learning optimal estimators for compressed sensing.For (1)\, I will pres
 ent an algorithm for score estimation based on piecewise polynomial regres
 sion\, yielding the first quasipolynomial-time algorithm for learning gene
 ral mixtures of Gaussians with polylogarithmically many components. For (2
 )\, I will give the first rigorous learning guarantee for algorithm unroll
 ing\, proving that a certain unrolled network\, when trained on compressed
  sensing examples\, learns to compete with Bayes approximate message passi
 ng.Based on joint works with Aayush Karan\, Vasilis Kontonis\, and Kulin S
 hah.\n&nbsp\;
LOCATION:External
END:VEVENT
END:VCALENDAR