Membership inference attack
PulseAugur coverage of Membership inference attack — every cluster mentioning Membership inference attack across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
Differential privacy fails to protect vulnerable subgroups in synthetic text releases
A new audit of differentially private synthetic text releases reveals that while differential privacy (DP) is effective at reducing average membership inference leakage, it disproportionately protects some records over …
-
New federated learning framework enhances privacy and robustness for sensitive data
Researchers have developed a new federated learning framework, DP-BR-FedAvg, designed to enhance security and privacy in cross-institutional model training for sectors like banking and healthcare. This framework integra…
-
New LAR method improves AI text detection and membership inference
Researchers have developed a new method called likelihood-array regression (LAR) to improve the detection of AI-generated text and identify if a specific text was used in training a language model. LAR evaluates token p…
-
New data poisoning method enhances VLM auditing
Researchers have developed MemCatalyst, a novel data poisoning technique designed to improve the effectiveness of data auditing for vision-language models (VLMs). This method uses two strategies, Poisoning Text (PT) and…
-
New method infers entity-level training data in LLMs
Researchers have introduced a new method for entity-level membership inference in large language models (LLMs). This approach aims to determine if information about a specific real-world entity, rather than just individ…
-
New research probes LLM inference, privacy, and code stylometry
Recent research explores the internal workings and security of large language models (LLMs). One study investigates how LLMs might form abstract representations similar to the human hippocampus to support inference, fin…
-
New framework audits synthetic AI data for privacy disclosures
Researchers have developed a new framework to audit synthetic data generated by AI models, aiming to detect and explain instances where private information from the training data might be leaked. The method distinguishe…
-
New research reveals how generative models retain training data signals
Researchers have identified a method to detect subtle traces of training data within generative models, even when the data isn't directly reproduced. By analyzing the interpolation path in Rectified Flows, they found a …
-
AI security research paper calls for more defense incentives
A recent paper published on arXiv highlights a significant imbalance in AI security research, with a disproportionate focus on attack methodologies over defensive strategies. The research indicates that attack papers ar…
-
LLM privacy study reveals context-dependent risks from various attacks
A new study published on arXiv investigates the privacy risks associated with large language models (LLMs) when used in interactive and retrieval-augmented systems. The research introduces a unified threat model and con…