• English
  • Deutsch
  • Log In
    or
  • Research Outputs
  • Projects
  • Researchers
  • Institutes
  • Statistics
Repository logo
Fraunhofer-Gesellschaft
  1. Home
  2. Fraunhofer-Gesellschaft
  3. Konferenzschrift
  4. Decoupling Autoencoders for Robust One-vs-Rest Classification
 
  • Details
  • Full
Options
2021
Conference Paper
Titel

Decoupling Autoencoders for Robust One-vs-Rest Classification

Abstract
One-vs-Rest (OVR) classification aims to distinguish a single class of interest from other classes. The concept of novelty detection and robustness to dataset shift becomes crucial in OVR when the scope of the rest class extends from the classes observed during training to unseen and possibly unrelated classes. In this work, we propose a novel architecture, namely Decoupling Autoencoder (DAE) to tackle the common issue of robustness w.r.t. out-of-distribution samples which is prevalent in classifiers such as multi-layer perceptrons (MLP) and ensemble architectures. Experiments on plain classification, outlier detection, and dataset shift tasks show DAE to achieve robust performance across these tasks compared to the baselines, which tend to fail completely, when exposed to dataset shift. W hile DAE and the baselines yield rather uncalibrated predictions on the outlier detection and dataset shift task, we found that DAE calibration is more stable across all tasks. Therefore, calibration measures applied to the classification task could also improve the calibration of the outlier detection and dataset shift scenarios for DAE.
Author(s)
Lübbering, Max
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS
Gebauer, Michael
TU Berlin
Ramamurthy, Rajkumar
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS
Bauckhage, Christian
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS
Sifa, Rafet
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS
Hauptwerk
IEEE 8th International Conference on Data Science and Advanced Analytics (DSAA 2021)
Konferenz
International Conference on Data Science and Advanced Analytics (DSAA) 2021
Thumbnail Image
DOI
10.1109/DSAA53316.2021.9564136
Language
English
google-scholar
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS
Tags
  • Training

  • Deep learning

  • Data science

  • Feature extraction

  • Robustness

  • Calibration

  • Risk management

  • One vs Rest Classific...

  • Outlier Detection

  • Dataset Shift

  • Cookie settings
  • Imprint
  • Privacy policy
  • Api
  • Send Feedback
© 2022