We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Blinder: End-to-end Privacy Protection in Sensing Systems via Personalized Federated Learning

Abstract: This paper proposes a sensor data anonymization model that is trained on decentralized data and strikes a desirable trade-off between data utility and privacy, even in heterogeneous settings where the collected sensor data have different underlying distributions. Our anonymization model, dubbed Blinder, is based on a variational autoencoder and discriminator networks trained in an adversarial fashion. We use the model-agnostic meta-learning framework to adapt the anonymization model trained via federated learning to each user's data distribution. We evaluate Blinder under different settings and show that it provides end-to-end privacy protection at the cost of increasing privacy loss by up to 4.00% and decreasing data utility by up to 4.24%, compared to the state-of-the-art anonymization model trained on centralized data. Our experiments confirm that Blinder can obscure multiple private attributes at once, and has sufficiently low power consumption and computational overhead for it to be deployed on edge devices and smartphones to perform real-time anonymization of sensor data.
Subjects: Machine Learning (cs.LG); Cryptography and Security (cs.CR)
Cite as: arXiv:2209.12046 [cs.LG]
  (or arXiv:2209.12046v2 [cs.LG] for this version)

Submission history

From: Xin Yang [view email]
[v1] Sat, 24 Sep 2022 16:22:43 GMT (2064kb,D)
[v2] Tue, 27 Sep 2022 03:36:42 GMT (2034kb,D)

Link back to: arXiv, form interface, contact.