Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This work re-evaluates Membership Inference Attacks (MIAs) for tabular data, revealing that existing MIA evaluations are biased towards high-dimensional settings and overlook critical factors in smaller datasets. Through a novel taxonomy and empirical study of four SOTA attacks across diverse tabular datasets and models, the authors find that class granularity and majority class prevalence significantly influence vulnerability, and that single-out records remain highly susceptible even with common defenses. The key result is that current MIA defenses offer only partial protection, failing to adequately safeguard the most vulnerable individuals.
Even with state-of-the-art defenses, individual records with unique signatures remain highly vulnerable to membership inference attacks in tabular data.
Membership Inference Attacks (MIAs) are currently the dominant approach to evaluating privacy in machine learning applications. Despite their significance, several concerns remain unexplored, particularly with regard to tabular data. Existing evaluations focus predominantly on high-dimensional datasets, leaving the privacy behavior of small-to-medium datasets unexplored. In this paper, we present a new taxonomy for MIAs and conduct a comprehensive study of the factors influencing vulnerability. We evaluate four distinct state-of-the-art attacks across diverse tabular datasets and model architectures. Our results show that datasets with higher class granularity and majority class percentage are more susceptible to MIAs. We further demonstrate that advanced MIAs are highly dependent on adversarial knowledge of the target model, while reference model-free attacks often achieve statistically comparable results. Most importantly, we show that records with unique signatures (single-outs) remain highly vulnerable, even when common defenses are applied. These findings indicate that current defenses reduce average leakage, but fail to adequately protect the most sensitive individuals.