No abstract

Machine learning (ML) models have been widely applied to various applications, including image classification, text generation, audio recognition, and graph data analysis. However, recent studies have shown that ML models are vulnerable to membership inference attacks (MIAs), which aim to infer whether a data record was used to train a target model or not. MIAs on ML models can directly lead to a privacy breach. For example, via identifying the fact that a clinical record that has been used to train a model associated with a certain disease, an attacker can infer that the owner of the clinical record has the disease with a high chance. In recent years, MIAs have been shown to be effective on various ML models, e.g., classification models and generative models. Meanwhile, many defense methods have been proposed to mitigate MIAs. Although MIAs on ML models form a newly emerging and rapidly growing research area, there has been no systematic survey on this topic yet. In this paper, we conduct the first comprehensive survey on membership inference attacks and defenses. We provide the taxonomies for both attacks and defenses, based on their characterizations, and discuss their pros and cons. Based on the limitations and gaps identified in this survey, we point out several promising future research directions to inspire the researchers who wish to follow this area. This survey not only serves as a reference for the research community but also provides a clear description for researchers outside this research domain. To further help the researchers, we have created an online resource repository, which we will keep updated with future relevant work. Interested readers can find the repository at https://github.com/HongshengHu/membership-inference-machine-learning-literature.

Current drift detection techniques detect a change in distribution within a stream. However, there are no current techniques that analyze the change in the rate of these detected changes. We coin the term stream volatility, to describe the rate of changes in a stream. A stream has a high volatility if changes are detected frequently and has a low volatility if changes are detected infrequently. We are particularly interested in a volatility shift which is a change in the rate of change (e.g. from high volatility to low volatility). We introduce and define the concept of stream volatility, and propose a novel technique to detect volatility on data streams in the presence of concept drifts. In the experiments we show our algorithm to be both fast and efficient. We also propose a new algorithm for drift detection called SEED that is faster and more memory efficient than the existing state-of-the-art drift detection approach. A faster drift detection algorithm has a flow-on benefit to the subsequent volatility detection stage because both algorithms run concurrently on the data stream.

No abstract

Recommender systems using Collaborative Filtering techniques are capable of make personalized predictions. However, these systems are highly vulnerable to profile injection attacks. Group attacks are attacks that target a group of items instead of one, and there are common attributes among these items. Such profiles will have a good probability of being similar to a large number of user profiles, making them hard to detect. We propose a novel technique for identifying group attack profiles which uses an improved metric based on Degree of Similarity with Top Neighbors (DegSim) and Rating Deviation from Mean Agreement (RDMA). We also extend our work with a detailed analysis of target item rating patterns. Experiments show that the combined methods can improve detection rates in user-based recommender systems.