Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May be Effective for Detecting Unknown Spurious Correlations

Item request has been placed! ×
Item request cannot be made. ×
loading   Processing Request
  • Additional Information
    • Publication Date:
      2022
    • Collection:
      Computer Science
    • Abstract:
      Recent work has suggested post-hoc explainers might be ineffective for detecting spurious correlations in Deep Neural Networks (DNNs). However, we show there are serious weaknesses with the existing evaluation frameworks for this setting. Previously proposed metrics are extremely difficult to interpret and are not directly comparable between explainer methods. To alleviate these constraints, we propose a new evaluation methodology, Explainer Divergence Scores (EDS), grounded in an information theory approach to evaluate explainers. EDS is easy to interpret and naturally comparable across explainers. We use our methodology to compare the detection performance of three different explainers - feature attribution methods, influential examples and concept extraction, on two different image datasets. We discover post-hoc explainers often contain substantial information about a DNN's dependence on spurious artifacts, but in ways often imperceptible to human users. This suggests the need for new techniques that can use this information to better detect a DNN's reliance on spurious correlations.
      Comment: Presented at the AIMLAI workshop at the 31st ACM International Conference on Information and Knowledge Management (CIKM 2022)
    • Accession Number:
      edsarx.2211.07650
  • Citations
    • ABNT:
      CARDOZO, S. et al. Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May be Effective for Detecting Unknown Spurious Correlations. [s. l.], 2022. Disponível em: https://ezproxy.mscc.edu/login?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsarx&AN=edsarx.2211.07650. Acesso em: 30 jan. 2023.
    • AMA 11th Edition:
      Cardozo S, Montero GI, Kazhdan D, et al. Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May be Effective for Detecting Unknown Spurious Correlations. 2022. Accessed January 30, 2023. https://ezproxy.mscc.edu/login?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsarx&AN=edsarx.2211.07650
    • APA 7th Edition:
      Cardozo, S., Montero, G. I., Kazhdan, D., Dimanov, B., Wijaya, M., Jamnik, M., & Lio, P. (2022). Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May be Effective for Detecting Unknown Spurious Correlations.
    • Chicago 17th Edition:
      Cardozo, Shea, Gabriel Islas Montero, Dmitry Kazhdan, Botty Dimanov, Maleakhi Wijaya, Mateja Jamnik, and Pietro Lio. 2022. “Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May Be Effective for Detecting Unknown Spurious Correlations.” https://ezproxy.mscc.edu/login?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsarx&AN=edsarx.2211.07650.
    • Harvard:
      Cardozo, S. et al. (2022) ‘Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May be Effective for Detecting Unknown Spurious Correlations’. Available at: https://ezproxy.mscc.edu/login?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsarx&AN=edsarx.2211.07650 (Accessed: 30 January 2023).
    • Harvard: Australian:
      Cardozo, S, Montero, GI, Kazhdan, D, Dimanov, B, Wijaya, M, Jamnik, M & Lio, P 2022, ‘Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May be Effective for Detecting Unknown Spurious Correlations’, viewed 30 January 2023, .
    • MLA 9th Edition:
      Cardozo, Shea, et al. Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May Be Effective for Detecting Unknown Spurious Correlations. 2022. EBSCOhost, ezproxy.mscc.edu/login?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsarx&AN=edsarx.2211.07650.
    • Chicago 17th Edition:
      Cardozo, Shea, Gabriel Islas Montero, Dmitry Kazhdan, Botty Dimanov, Maleakhi Wijaya, Mateja Jamnik, and Pietro Lio. “Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May Be Effective for Detecting Unknown Spurious Correlations,” 2022. https://ezproxy.mscc.edu/login?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsarx&AN=edsarx.2211.07650.
    • Vancouver/ICMJE:
      Cardozo S, Montero GI, Kazhdan D, Dimanov B, Wijaya M, Jamnik M, et al. Explainer Divergence Scores (EDS): Some Post-Hoc Explanations May be Effective for Detecting Unknown Spurious Correlations. 2022 [cited 2023 Jan 30]; Available from: https://ezproxy.mscc.edu/login?url=https://search.ebscohost.com/login.aspx?direct=true&site=eds-live&db=edsarx&AN=edsarx.2211.07650