PuSH - Publikationsserver des Helmholtz Zentrums München

Bani-Harouni, D.* ; Mueller, T.T.* ; Rueckert, D.* ; Kaissis, G.

Gradient Self-alignment in Private Deep Learning.

In: (26th International Conference on Medical Image Computing and Computer-Assisted Intervention , MICCAI 2023, 8 - 12 October 2023, Vancouver, CANADA). Berlin [u.a.]: Springer, 2023. 89-97 (Lect. Notes Comput. Sc. ; 14393)
DOI
Open Access Green möglich sobald Postprint bei der ZB eingereicht worden ist.
Differential Privacy (DP) has become a gold-standard to preserve privacy in deep learning. Intuitively speaking, DP ensures that the output of a model is approximately invariant to the inclusion or exclusion of a single individual’s data from the training set. There is, however, a trade-off between privacy and utility. DP models tend to perform worse than non-DP models trained on the same data. This is caused by the clipping of per-sample gradients and the addition of noise required for DP guarantees causing an obfuscation of the individual data point’s contribution. In this work, we propose a method to reduce this discrepancy by improving the alignment between the per-sample gradients of each individual training sample with its non-DP gradient by increasing their cosine similarity. Optimizing the alignment in only a relevant subset of gradient dimensions, further improves the performance. We evaluate our method on CIFAR-10 and a pediatric pneumonia chest x-ray dataset.
Altmetric
Weitere Metriken?
[➜Einloggen]
Zusatzinfos bearbeiten [➜Einloggen]
Publikationstyp Artikel: Konferenzbeitrag
Korrespondenzautor
Schlagwörter Differential Privacy ; Gradient Alignment ; Private Learning
ISSN (print) / ISBN 0302-9743
e-ISSN 1611-3349
Konferenztitel 26th International Conference on Medical Image Computing and Computer-Assisted Intervention , MICCAI 2023
Konferzenzdatum 8 - 12 October 2023
Konferenzort Vancouver, CANADA
Zeitschrift Lecture Notes in Computer Science
Quellenangaben Band: 14393, Heft: , Seiten: 89-97 Artikelnummer: , Supplement: ,
Verlag Springer
Verlagsort Berlin [u.a.]
Nichtpatentliteratur Publikationen
Institut(e) Institute for Machine Learning in Biomed Imaging (IML)