Hauptmenü
  • Autor
    • Malle, Bernd
    • Kieseberg, Peter
    • Holzinger, Andreas
  • TitelInteractive Anonymization for Privacy aware Machine Learning
  • Datei
  • Persistent Identifier
  • Erscheinungsjahr2017
  • Seiten15
  • LicenceCC BY-ND
  • ZugriffsrechteCC-BY
  • Konferenz NameEuropean Conference on Machine Learning and Knowledge Discovery ECML-PKDD
  • Konferenz OrtSkopje
  • Konferenz StaatMazedonien, ehemalige jugoslawische Republik
  • Konferenz URLhttp://ecmlpkdd2017.ijs.si/
  • Download Statistik242
  • Peer ReviewJa
  • AbstractPrivacy aware Machine Learning is the discipline of applying Machine Learning techniques in such a way as to protect and retain personal identities during the process. This is most easily achieved by first anonymizing a dataset before releasing it for the purpose of data mining or knowledge extraction. Starting in June 2018, this will also remain the sole legally permitted way within the EU to release data without granting people involved the right to be forgotten, i.e. the right to have their data deleted on request. To governments, organizations and corporations, this represents a serious impediment to research operations, since any anonymization results in a certain degree of reduced data utility. In this paper we propose applying human background knowledge via interactive Machine Learning to the process of anonymization; this is done by eliciting human preferences for preserving some attribute values over others in the light of specific tasks. Our experiments show that human knowledge can yield measurably better classification results than a rigid automatic approach. However, the impact of interactive learning in the field of anonymization will largely depend on the experimental setup, such as an appropriate choice of application domain as well as suitable test subjects.