The right to be forgotten states that a data owner has the right to erase their data from an entity storing it. In the context of machine learning (ML), the right to be forgotten requires an ML model owner to remove the data owner's data from the training set used to build the ML model, a process known asmachine unlearning. While originally designed to protect the privacy of the data owner, we argue that machine unlearning may leave some imprint of the data in the ML model and thus create unintended privacy risks. In this paper, we perform the first study on investigating the unintended information leakage caused by machine unlearning. We propose a novel membership inference attack that leverages the different outputs of an ML model's two versions to infer whether a target sample is part of the training set of the original model but out of the training set of the corresponding unlearned model. Our experiments demonstrate that the proposed membership inference attack achieves strong performance. More importantly, we show that our attack in multiple cases outperforms the classical membership inference attack on the original ML model, which indicates that machine unlearning can have counterproductive effects on privacy. We notice that the privacy degradation is especially significant for well-generalized ML models where classical membership inference does not perform well. We further investigate four mechanisms to mitigate the newly discovered privacy risks and show that releasing the predicted label only, temperature scaling, and differential privacy are effective. We believe that our results can help improve privacy protection in practical implementations of machine unlearning. \footnoteOur code is available at \urlhttps://github.com/MinChen00/UnlearningLeaks.

This systematic review evaluates recent literature from January 2021 to March 2024 on large language model (LLM) applications across diverse medical specialties. Searching PubMed, Web of Science, and Scopus, we included 84 studies. LLMs were applied to tasks such as clinical natural language processing, medical decision support, education, and aiding diagnostic processes. While studies reported benefits such as improved efficiency and, in some specific NLP tasks, high accuracy above 90%, significant challenges persist concerning reliability, ethical implications, and performance consistency, with accuracy in broader diagnostic support applications showing substantial variability, with some as low as 3%. The overall risk of bias in the reviewed literature was considerably low in 72 studies. Key findings highlight a substantial heterogeneity in LLM performance across different medical tasks and contexts, preventing meta-analysis due to a lack of standardized methodologies. Future efforts should prioritize developing domain-specific LLMs using robust medical data and establishing rigorous validation standards to ensure their safe and effective clinical integration. Trial registration: PROSPERO (CRD42024561381).

Graph-based fraud detection has attracted increasing attention in recent years, reflecting its growing potential in mitigating sophisticated fraudulent activities. The main objective of graph-based fraud detection is to discern between fraud-sters and normal entities within graphs. As fraudsters adopt increasingly sophisticated camouflage tactics, combating them has become an urgent task. Despite the complex interactions within real-world networks involving high-order relations, ex-isting graph-based fraud detection methods often neglect non-pairwise relationships among entities in graphs. Thus, we empha-size the significance of investigating beyond pairwise relationships for building an effective fraud detection model. In this paper, we propose constructing a hypergraph from the original input graph to encapsulate comprehensive high-order relations and present TROPICAL, a novel TRansfOrmer-based hyPergraph LearnIng for detecting CAmouflaged maLicious actors in online social networks. TROPICAL learns representations by processing different hyperedge groups and incorporates positional encodings into the aggregated information to enhance their distinctiveness. Subsequently, the model feeds the learned aggregated sequential information into the transformer encoder, achieving rich rep-resentations for effective camouflaged fraudster detection. The superiority of TROPICAL is demonstrated through experiments conducted on two real-world datasets, compared against the state-of-the-art fraud detection models. The source codes and datasets of our work are available at https://github.comNenusHaghighi/TROPICAL.