BEGIN:VCALENDAR
VERSION:2.0
CALSCALE:GREGORIAN
PRODID:UW-Madison-Physics-Events
BEGIN:VEVENT
SEQUENCE:0
UID:UW-Physics-Event-6177
DTSTART:20201104T170000Z
DTEND:20201104T181500Z
DTSTAMP:20260414T143216Z
LAST-MODIFIED:20201025T033720Z
LOCATION:Online Seminar: Please sign up for our mailing list at www.ph
 ysicsmeetsml.org for zoom link
SUMMARY:Flow-based likelihoods for non-Gaussian inference\, Physics âˆ
 © ML Seminar\, Ana Diaz Rivero\, Harvard University
DESCRIPTION:We investigate the use of data-driven likelihoods to bypas
 s a key assumption made in many scientific analyses\, which is that th
 e true likelihood of the data is Gaussian. In particular\, we suggest 
 using the optimization targets of flow-based generative models\, a cla
 ss of models that can capture complex distributions by transforming a 
 simple base distribution through layers of nonlinearities. We call the
 se flow-based likelihoods (FBL). We analyze the accuracy and precision
  of the reconstructed likelihoods on mock Gaussian data\, and show tha
 t simply gauging the quality of samples drawn from the trained model i
 s not a sufficient indicator that the true likelihood has been learned
 . We nevertheless demonstrate that the likelihood can be reconstructed
  to a precision equal to that of sampling error due to a finite sample
  size. We then apply FBLs to mock weak lensing convergence power spect
 ra\, a cosmological observable that is significantly non-Gaussian (NG)
 . We find that the FBL captures the NG signatures in the data extremel
 y well\, while other commonly-used data-driven likelihoods\, such as G
 aussian mixture models and independent component analysis\, fail to do
  so. This suggests that works that have found small posterior shifts i
 n NG data with data-driven likelihoods such as these could be underest
 imating the impact of non-Gaussianity in parameter constraints. By int
 roducing a suite of tests that can capture different levels of NG in t
 he data\, we show that the success or failure of traditional data-driv
 en likelihoods can be tied back to the structure of the NG in the data
 . Unlike other methods\, the flexibility of the FBL makes it successfu
 l at tackling different types of NG simultaneously. Because of this\, 
 and consequently their likely applicability across datasets and domain
 s\, we encourage their use for inference when sufficient mock data are
  available for training.
URL:https://www.physics.wisc.edu/events/?id=6177
END:VEVENT
END:VCALENDAR