[Theory] TODAY: [TTIC Talks] 10/10 Research at TTIC: Mahdi Haghifam, TTIC

[Brandie Jones via Theory]

*When:         *October 10th *at 11am CT  *


*Where:*        Talk will be given *live, in-person* at

                       TTIC, 6045 S. Kenwood Avenue

                        5th Floor, Room 530


*Virtually:*    via Panopto (Livestream
<https://uchicago.hosted.panopto.com/Panopto/Pages/Viewer.aspx?id=ee622f4c-8f38-431e-b637-b31a00f2af30>
)

*Who:*            Mahdi Haghifam, TTIC

*Title:*             The Sample Complexity of Membership Inference and
Privacy Auditing

*Abstract:*     Membership Inference Attacks (MIAs) aim to determine if it
is possible, using only a model's output, to distinguish training data from
held-out samples. MIAs are an influential tool in machine learning privacy,
with deep connections to measuring memorization, auditing differentially
private algorithms, understanding generalization, and establishing
fundamental privacy lower bounds. The success of these attacks often hinges
on the attacker's background knowledge of the underlying data distribution.
A key question of this talk is: how much knowledge about the underlying
data distribution is actually necessary for an attack to succeed?

This talk makes this question precise using the language of sample
complexity. We investigate a practical scenario where the attacker must
learn about the data distribution from a limited number of auxiliary
samples. Focusing on the fundamental setting of Gaussian mean estimation,
we reveal a sharp divide based on the attacker's prior knowledge. An
attacker who knows the data's covariance structure requires only a linear
number of auxiliary samples to mount a successful attack. Without this
knowledge, however, the required number of samples becomes at least
quadratic. This result highlights that an adversary's prior knowledge is a
critical factor in a model's privacy risk and has significant implications
for how we design and audit private machine learning systems.
The talk is based on a joint work with Adam Smith and Jon Ullman and can be
found at https://arxiv.org/abs/2508.19458.
<https://arxiv.org/abs/2508.19458>



***********************************************************************************************

*Masks are optional in all common areas. **Full visitor guidance is
available at ttic.edu/visitors <http://ttic.edu/visitors>.*

***********************************************************************************************

*Research at TTIC Seminar Series*



TTIC is hosting a weekly seminar series presenting the research currently
underway at the Institute. Every week a different TTIC faculty member will
present their research.  The lectures are intended for students
seeking research topics and advisors, and for the general TTIC and
University of Chicago communities interested in hearing what their
colleagues are up to.



*Brandie Jones *
*Executive **Administrative Assistant*
*Outreach Administrator *
Toyota Technological Institute
6045 S. Kenwood Avenue
Chicago, IL  60637
www.ttic.edu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20251010/b83fa503/attachment.html>