Membership Inference Attacks
PulseAugur coverage of Membership Inference Attacks — every cluster mentioning Membership Inference Attacks across labs, papers, and developer communities, ranked by signal.
1 day(s) with sentiment data
-
New GraphToxin Attack Reconstructs Deleted Data from GNNs
Researchers have developed GraphToxin, a novel attack capable of reconstructing entire unlearned graphs from graph neural networks (GNNs). This method goes beyond previous membership inference attacks by not only recove…
-
New framework Adaptive Diffusion Freezing enhances privacy in diffusion models
Researchers have introduced Adaptive Diffusion Freezing (ADF), a new framework designed to enhance privacy in diffusion models against membership inference attacks (MIAs). This method utilizes cross-timestep adaptive fr…
-
New study benchmarks privacy risks in NLP text classifiers
A new study on arXiv evaluates the privacy risks associated with training natural language processing (NLP) text classifiers. Researchers benchmarked membership inference attacks (MIAs) on the GLUE SST-2 sentiment datas…
-
New method predicts AI model privacy leakage using spectral analysis
Researchers have developed a method to predict privacy leakage from machine learning models using spectral metrics derived from their weights. This approach aims to bypass the need for computationally expensive shadow m…
-
New method quantifies privacy risk in synthetic tabular data
Researchers have developed a new method to quantify the risk of membership disclosure in tabular synthetic data using kernel density estimators (KDEs). This approach models the distances between synthetic and training d…
-
New CDML method enhances privacy and accuracy in continual gait identification
Researchers have developed Code Division Modulation Layers (CDML) to address challenges in continual learning for biometric identification systems, specifically gait identification. This new approach aims to maintain hi…
-
New framework unifies membership inference attacks across generative models
Researchers have developed a unified framework for membership inference attacks (MIA) that can be applied across various generative model modalities, including text-to-text, text-to-image, and image-to-text. This new ap…
-
New AI Cyber Attack Type Targets Patient Privacy
A new category of cyber attack, known as Membership Inference Attacks (MIAs), has been identified as a threat to patient privacy. These attacks specifically target individuals within medical datasets, potentially exposi…
-
New attack reveals privacy risks in tabular foundation models
Researchers have identified significant privacy vulnerabilities in tabular foundation models, particularly within their attention layers. A new attack, AMIA, leverages transformer attention patterns to effectively perfo…
-
New benchmarks tackle privacy risks in large language models
Researchers have developed new methods to evaluate membership inference attacks (MIAs) against large language models (LLMs), particularly focusing on audio and text modalities. The first study introduces a systematic ev…
-
Researchers propose $\mu \approx \varepsilon/5$ conversion for Gaussian differential privacy
Researchers have published a paper detailing methods for converting privacy parameters between pure differential privacy ($\varepsilon$) and Gaussian differential privacy (GDP, $\mu$). The study proposes principled mapp…
-
Counterfactuals pose privacy risks, new research shows
Researchers have demonstrated that counterfactual explanations, used to clarify machine learning model decisions, can be exploited for privacy attacks. By adapting methods developed for synthetic data, these attacks can…
-
Bayesian sampling enhances AI model privacy attack efficiency
Researchers have developed a new method called Bayesian Membership Inference Attack (BMIA) to more efficiently detect if specific data points were used in a model's training. This approach uses Bayesian sampling and Lap…
-
New research refines evaluation of AI model privacy attacks
Researchers are developing new frameworks and methods to evaluate the effectiveness and reliability of membership inference attacks (MIAs), which are used to detect if specific data was used in training machine learning…
-
New attack method breaches privacy of AI safety classifiers
Researchers have developed a new method to attack the privacy of safety classifiers used in generative AI systems. These classifiers, trained on sensitive data like discussions of self-harm, are vulnerable to membership…