TR2021-016

AutoBayes: Automated Bayesian Graph Exploration for Nuisance-Robust Inference


    •  Demir, A., Koike-Akino, T., Wang, Y., Erdogmus, D., "AutoBayes: Automated Bayesian Graph Exploration for Nuisance-Robust Inference", IEEE Access, DOI: 10.1109/​ACCESS.2021.3064530, Vol. 9, pp. 39955-39972, March 2021.
      BibTeX TR2021-016 PDF
      • @article{Demir2021mar,
      • author = {Demir, Andac and Koike-Akino, Toshiaki and Wang, Ye and Erdogmus, Deniz},
      • title = {AutoBayes: Automated Bayesian Graph Exploration for Nuisance-Robust Inference},
      • journal = {IEEE Access},
      • year = 2021,
      • volume = 9,
      • pages = {39955--39972},
      • month = mar,
      • doi = {10.1109/ACCESS.2021.3064530},
      • issn = {2169-3536},
      • url = {https://www.merl.com/publications/TR2021-016}
      • }
  • MERL Contacts:
  • Research Areas:

    Artificial Intelligence, Machine Learning, Robotics, Signal Processing

Abstract:

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