How Do Input Attributes Impact the Privacy Loss in Differential Privacy?
Not on the map. Why?
Abstract (arXiv)
Differential privacy (DP) is typically formulated as a worst-case privacy guarantee over all individuals in a database. More recently, extensions to individual subjects or their attributes, have been introduced. Under the individual/per-instance DP interpretation, we study the connection between the per-subject gradient norm in DP neural networks and individual privacy loss and introduce a novel metric termed the Privacy Loss-Input Susceptibility (PLIS), which allows one to apportion the subject's privacy loss to their input attributes. We experimentally show how this enables the identification of sensitive attributes and of subjects at high risk of data reconstruction.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 10 of its 19 references that are in the map. First posted on arXiv 18 Nov 2022.
Builds on
10 of 19 references.
References · 10
- 2020Inverting Gradients -- How easy is it to break privacy in federated learning?cited by 707
- 2017Renyi Differential Privacycited by 643
- 2020A Survey on Neural Network Interpretabilitycited by 291
- 2022Reconstructing Training Data with Informed Adversariescited by 143
- 2016A Haar Wavelet-Based Perceptual Similarity Index for Image Quality Assessmentcited by 57
- 2020Individual Privacy Accounting via a Renyi Filtercited by 57
- 2022Bounding Training Data Reconstruction in Private (Deep) Learningcited by 43
- 2021Measuring Data Leakage in Machine-Learning Models with Fisher Informationcited by 31
- 2020Layer-wise Characterization of Latent Information Leakage in Federated Learningcited by 19
- 2019Element Level Differential Privacy: The Right Granularity of Privacycited by 6
Built on by
Citing papers per year: 2023: 2.
2 papers.