Evaluating the Impact of Local Differential Privacy on Utility Loss via Influence Functions. (arXiv:2309.08678v1 [cs.LG])

Abstract:How to properly set the privacy parameter in differential privacy (DP) has been an open question in DP research since it was first proposed in 2006. In this work, we demonstrate the ability of influence functions to offer insight into how a specific privacy parameter value will affect a model's test loss in the randomized response-based local DP setting. Our

Read more

Related Posts

• Spectral-DP: Differentially Private Deep Learning through Spectral Perturbation and Filtering. (arXiv:2307.13231v1 [cs.LG])a
• A Neighbourhood-Aware Differential Privacy Mechanism for Static Word Embeddings. (arXiv:2309.10551v1 [cs.LG])a
• Poster: Patient Community — A Test Bed For Privacy Threat Analysis. (arXiv:2308.02272v1 [cs.CR])a
• Local Differential Privacy in Graph Neural Networks: a Reconstruction Approach. (arXiv:2309.08569v1 [cs.LG])a
• FairDP: Certified Fairness with Differential Privacy. (arXiv:2305.16474v2 [cs.LG] UPDATED)a