Correcting Predictions for Approximate Bayesian Inference

Tomasz Kusmierczyk, Joseph Sakaya, Arto Klami

Forskningsoutput: Kapitel i bok/rapport/konferenshandlingKonferensbidragVetenskapligPeer review


Bayesian models quantify uncertainty and facilitate optimal decision-making in downstream applications. For most models, however, practitioners are forced to use approximate inference techniques that lead to sub-optimal decisions due to incorrect posterior predictive distributions. We present a novel approach that corrects for inaccuracies in posterior inference by altering the decision-making process. We train a separate model to make optimal decisions under the approximate posterior, combining interpretable Bayesian modeling with optimization of direct predictive accuracy in a principled fashion. The solution is generally applicable as a plug-in module for predictive decision-making for arbitrary probabilistic programs, irrespective of the posterior inference strategy. We demonstrate the approach empirically in several problems, confirming its potential.
Titel på gästpublikationThe thirty-fourth AAAI conference on artificial intelligence
Antal sidor8
FörlagAssociation for the Advancement of Artificial Intelligence (AAAI)
Utgivningsdatumfeb 2020
ISBN (elektroniskt)978-1-57735-835-0
StatusPublicerad - feb 2020
MoE-publikationstypA4 Artikel i en konferenspublikation
Evenemang34th AAAI Conference on Artificial Intelligence (AAAI-20) - New York, Förenta Staterna (USA)
Varaktighet: 7 feb 202012 feb 2020


NamnProceedings of the AAAI Conference on Artificial Intelligence
FörlagAAAI Press
ISSN (tryckt)2159-5399
ISSN (elektroniskt)2374-3468


  • 113 Data- och informationsvetenskap

Citera det här