The broad availability of computational resources and the recent scientific progresses made deep learning the elected class of algorithms to solve complex tasks. Besides their deployment, two problems have risen: fighting biases in data and privacy preservation of sensitive attributes. Many solutions have been proposed, some of which deepen their roots in the pre-deep learning theory. There are many similarities between debiasing and privacy preserving approaches: how far apart are these two worlds, when the private information overlaps the bias?In this work we investigate the possibility of deploying debiasing strategies also to prevent privacy leakage. In particular, empirically testing on state-of-the-art datasets, we observe that there exists a subset of debiasing approaches which are also suitable for privacy preservation. We identify as the discrimen the capability of effectively hiding the biased information, rather than simply re-weighting it

Bridging the gap between debiasing and privacy for deep learning

Barbano, Carlo Alberto;Tartaglione, Enzo;Grangetto, Marco
2021

Abstract

The broad availability of computational resources and the recent scientific progresses made deep learning the elected class of algorithms to solve complex tasks. Besides their deployment, two problems have risen: fighting biases in data and privacy preservation of sensitive attributes. Many solutions have been proposed, some of which deepen their roots in the pre-deep learning theory. There are many similarities between debiasing and privacy preserving approaches: how far apart are these two worlds, when the private information overlaps the bias?In this work we investigate the possibility of deploying debiasing strategies also to prevent privacy leakage. In particular, empirically testing on state-of-the-art datasets, we observe that there exists a subset of debiasing approaches which are also suitable for privacy preservation. We identify as the discrimen the capability of effectively hiding the biased information, rather than simply re-weighting it
IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)
virtual
October 11-17, 2021
2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)
IEEE
3799
3808
978-1-6654-0191-3
https://openaccess.thecvf.com/content/ICCV2021W/RPRMI/papers/Barbano_Bridging_the_Gap_Between_Debiasing_and_Privacy_for_Deep_Learning_ICCVW_2021_paper.pdf
Barbano, Carlo Alberto; Tartaglione, Enzo; Grangetto, Marco
File in questo prodotto:
File Dimensione Formato  
Bridging_the_gap_between_debiasing_and_privacy_for_deep_learning.pdf

non disponibili

Tipo di file: PDF EDITORIALE
Dimensione 2.29 MB
Formato Adobe PDF
2.29 MB Adobe PDF   Visualizza/Apri   Richiedi una copia

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/2318/1825746
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact