WEBINARS IN STATISTICS @ COLLEGIO CARLO ALBERTO https://www.carloalberto.org/events/category/seminars/seminars-in-statistics/
Joint initiative with
MIDAS COMPLEX MODELING RESEARCH NETWORK http://midas.mat.uc.cl/network
Giovedi 17 Dicembre 2020, alle ore 17:00, si terrà il seguente webinar:
------------------------------------------------
Speaker: *David Rossell *(Universitat Pompeu Fabra, Barcelona, Spain)
Title: *Approximate Laplace approximation*
Zoom link:
https://us02web.zoom.us/j/84312531120?pwd=VVJraStLVkR2M0w0ZXZnU3M0MFp2UT09 https://us02web.zoom.us/j/88252069649?pwd=V2Z3b1UrZVVWNWZ4OXhydUtIakxpUT09
Meeting ID: 843 1253 1120
Passcode: 105560
Abstract:
Bayesian model selection requires an integration exercise in order to assign posterior model probabilities to each candidate model. The computation becomes cumbersome when the integral has no closed-form, particularly when the sample size is large, or the number of models is large. We present a simple yet powerful idea based on the Laplace approximation (LA) to an integral. LA uses a quadratic Taylor expansion at the mode of the integrand and is typically quite accurate, but requires cumbersome likelihood evaluations (for large n) an optimization (for large p). We propose the approximate Laplace approximation (ALA), which uses an Taylor expansion at the null parameter value. ALA brings very significant speed-ups by avoiding optimizations altogether, and evaluating likelihoods via sufficient statistics. ALA is an approximate inference method equipped with strong model selection properties in the family of non-linear GLMs, attaining comparable rates to exact computation. When (inevitably) the model is misspecified the ALA rates can actually be faster than for exact computation, depending on the type of misspecification. We show examples in non-linear Gaussian regression with non-local priors, for which no closed-form integral exists, as well as non-linear logistic, Poisson and survival regression.
------------------------------------------------
Il webinar è organizzato dalla "de Castro" Statistics Initiative
www.carloalberto.org/stats
in collaborazione con il Collegio Carlo Alberto e rientra nel Complex Data Modeling Research Network
midas.mat.uc.cl/network
Cordiali saluti,
Pierpaolo De Blasi
---
University of Torino & Collegio Carlo Alberto
carloalberto.org/pdeblasi https://sites.google.com/a/carloalberto.org/pdeblasi/
Scusate per l'inconveniente, l'annuncio del seminario di David Rossell di oggi pomeriggio, vedi messaggio in calce, conteneva un errore nel link a Zoom, che si risolve 1) facendo copia e incolla direttamente sul browser oppure 2) utilizzando il seguente link
https://us02web.zoom.us/j/84312531120?pwd=VVJraStLVkR2M0w0ZXZnU3M0MFp2UT09
Pierpaolo De Blasi
---------- Forwarded message --------- Da: Pierpaolo De Blasi pierpaolo.deblasi@unito.it Date: lun 14 dic 2020 alle ore 12:33 Subject: Webinar DAVID ROSSELL To: random@mail.dm.unipi.it
WEBINARS IN STATISTICS @ COLLEGIO CARLO ALBERTO https://www.carloalberto.org/events/category/seminars/seminars-in-statistics/
Joint initiative with
MIDAS COMPLEX MODELING RESEARCH NETWORK http://midas.mat.uc.cl/network
Giovedi 17 Dicembre 2020, alle ore 17:00, si terrà il seguente webinar:
------------------------------------------------
Speaker: *David Rossell *(Universitat Pompeu Fabra, Barcelona, Spain)
Title: *Approximate Laplace approximation*
Zoom link:
https://us02web.zoom.us/j/84312531120?pwd=VVJraStLVkR2M0w0ZXZnU3M0MFp2UT09 https://us02web.zoom.us/j/88252069649?pwd=V2Z3b1UrZVVWNWZ4OXhydUtIakxpUT09
Meeting ID: 843 1253 1120
Passcode: 105560
Abstract:
Bayesian model selection requires an integration exercise in order to assign posterior model probabilities to each candidate model. The computation becomes cumbersome when the integral has no closed-form, particularly when the sample size is large, or the number of models is large. We present a simple yet powerful idea based on the Laplace approximation (LA) to an integral. LA uses a quadratic Taylor expansion at the mode of the integrand and is typically quite accurate, but requires cumbersome likelihood evaluations (for large n) an optimization (for large p). We propose the approximate Laplace approximation (ALA), which uses an Taylor expansion at the null parameter value. ALA brings very significant speed-ups by avoiding optimizations altogether, and evaluating likelihoods via sufficient statistics. ALA is an approximate inference method equipped with strong model selection properties in the family of non-linear GLMs, attaining comparable rates to exact computation. When (inevitably) the model is misspecified the ALA rates can actually be faster than for exact computation, depending on the type of misspecification. We show examples in non-linear Gaussian regression with non-local priors, for which no closed-form integral exists, as well as non-linear logistic, Poisson and survival regression.
------------------------------------------------
Il webinar è organizzato dalla "de Castro" Statistics Initiative
www.carloalberto.org/stats
in collaborazione con il Collegio Carlo Alberto e rientra nel Complex Data Modeling Research Network
midas.mat.uc.cl/network
Cordiali saluti,
Pierpaolo De Blasi
---
University of Torino & Collegio Carlo Alberto
carloalberto.org/pdeblasi https://sites.google.com/a/carloalberto.org/pdeblasi/
-
Pierpaolo De Blasi