Generating Fake Data to Fake Privacy Pryers
Keywords: Privacy, Data valuation, Generative models
Abstract: Asymmetry of data complexity and model capacity can create privacy vulnerability. That is because if there are relatively fewer data points while the model capacity is relatively higher, a model may memorize almost all the data points. As a remedy for the issue, more data samples can be generated. When generating more data samples, the aim is to protect and promote the original data as privacy-safe as possible while generating more privacy-risky data samples to fake privacy attackers. To enable the aim, we investigate each individual data sample's privacy level, unlike existing studies that only take into account an overall dataset's privacy, which is not precisely effective. We show how effective our generative approach is in combating privacy attacks. Our work is novel in that we propose a sample-level valuation, and data transformation and generation approach in the privacy domain.
Supplementary Material: zip
Primary Area: alignment, fairness, safety, privacy, and societal considerations
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5200
Loading