eduzhai > Applied Sciences > Engineering >

AutoBayes Automated Bayesian Graph Exploration for Nuisance-Robust Inference

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 24 pages

Abstract: Learning data representations that capture task-related features, but areinvariant to nuisance variations remains a key challenge in machine learning.We introduce an automated Bayesian inference framework, called AutoBayes, thatexplores different graphical models linking classifier, encoder, decoder,estimator and adversarial network blocks to optimize nuisance-invariant machinelearning pipelines. AutoBayes also enables learning disentangledrepresentations, where the latent variable is split into multiple pieces toimpose various relationships with the nuisance variation and task labels. Webenchmark the framework on several public datasets, and provide analysis of itscapability for subject-transfer learning with without variational modeling andadversarial training. We demonstrate a significant performance improvement withensemble learning across explored graphical models.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×