1-7hit |
In this paper, we propose a selective membership inference attack method that determines whether certain data corresponding to a specific class are being used as training data for a machine learning model or not. By using the proposed method, membership or non-membership can be inferred by generating a decision model from the prediction of the inference models and training the confidence values for the data corresponding to the selected class. We used MNIST as an experimental dataset and Tensorflow as a machine learning library. Experimental results show that the proposed method has a 92.4% success rate with 5 inference models for data corresponding to a specific class.
Zhikai XU Hongli ZHANG Xiangzhan YU Shen SU
Location-based services (LBSs) are useful for many applications in internet of things(IoT). However, LBSs has raised serious concerns about users' location privacy. In this paper, we propose a new location privacy attack in LBSs called hidden location inference attack, in which the adversary infers users' hidden locations based on the users' check-in histories. We discover three factors that influence individual check-in behaviors: geographic information, human mobility patterns and user preferences. We first separately evaluate the effects of each of these three factors on users' check-in behaviors. Next, we propose a novel algorithm that integrates the above heterogeneous factors and captures the probability of hidden location privacy leakage. Then, we design a novel privacy alert framework to warn users when their sharing behavior does not match their sharing rules. Finally, we use our experimental results to demonstrate the validity and practicality of the proposed strategy.
Chittaphone PHONHARATH Kenji HASHIMOTO Hiroyuki SEKI
We study a static analysis problem on k-secrecy, which is a metric for the security against inference attacks on XML databases. Intuitively, k-secrecy means that the number of candidates of sensitive data of a given database instance or the result of unauthorized query cannot be narrowed down to k-1 by using available information such as authorized queries and their results. In this paper, we investigate the decidability of the schema k-secrecy problem defined as follows: for a given XML database schema, an authorized query and an unauthorized query, decide whether every database instance conforming to the given schema is k-secret. We first show that the schema k-secrecy problem is undecidable for any finite k>1 even when queries are represented by a simple subclass of linear deterministic top-down tree transducers (LDTT). We next show that the schema ∞-secrecy problem is decidable for queries represented by LDTT. We give an algorithm for deciding the schema ∞-secrecy problem and analyze its time complexity. We show the schema ∞-secrecy problem is EXPTIME-complete for LDTT. Moreover, we show similar results LDTT with regular look-ahead.
Kenji HASHIMOTO Hiroto KAWAI Yasunori ISHIHARA Toru FUJIWARA
This paper discusses verification of the security against inference attacks on XML databases in the presence of a functional dependency. So far, we have provided the verification method for k-secrecy, which is a metric for the security against inference attacks on databases. Intuitively, k-secrecy means that the number of candidates of sensitive data (i.e., the result of unauthorized query) of a given database instance cannot be narrowed down to k-1 by using available information such as authorized queries and their results. In this paper, we consider a functional dependency on database instances as one of the available information. Functional dependencies help attackers to reduce the number of the candidates for the sensitive information. The verification method we have provided cannot be naively extended to the k-secrecy problem with a functional dependency. The method requires that the candidate set can be captured by a tree automaton, but the candidate set when a functional dependency is considered cannot be always captured by any tree automaton. We show that the ∞-secrecy problem in the presence of a functional dependency is decidable when a given unauthorized query is represented by a deterministic topdown tree transducer, without explicitly computing the candidate set.
Kenji HASHIMOTO Kimihide SAKANO Fumikazu TAKASUKA Yasunori ISHIHARA Toru FUJIWARA
This paper discusses verification of the security against inference attacks on XML databases. First, a security definition called k-secrecy against inference attacks on XML databases is proposed. k-secrecy with an integer k > 1 (or k = ∞) means that attackers cannot narrow down the candidates for the value of the sensitive information to k - 1 (or finite), using the results of given authorized queries and schema information. Secondly, an XML query model such that verification can be performed straightforwardly according to the security definition is presented. The query model can represent practical queries which extract some nodes according to any of their neighboring nodes such as ancestors, descendants, and siblings. Thirdly, another refinement of the verification method is presented, which produces much smaller intermediate results if a schema contains no arbitrarily recursive element. The correctness of the refinement is proved, and the effect of the refinement in time and space efficiency has been confirmed by experiment.
Yasunori ISHIHARA Shuichiro AKO Toru FUJIWARA
Inference attacks mean that a user derives information on the execution results of unauthorized queries from the execution results of authorized queries. Most of the studies on inference attacks so far have focused on only inference of positive information (i.e., what value is the execution result of a given unauthorized query). However, negative information (i.e., what value is never the execution result of a given unauthorized query) is also sensitive in many cases. This paper presents the following results on the security against inference attacks on negative information in object-oriented databases. First, inference of negative information is formalized under a model of object-oriented databases called method schemas. Then, the following two types of security problems are defined: (1) Is a given database instance secure against inference attacks on given negative information? (2) Are all of the database instances of a given database schema secure against inference attacks on given negative information? It is shown that the first problem is decidable in polynomial time in the description size of the database instance while the second one is undecidable. A decidable sufficient condition for any database instance of a given database schema to be secure is also proposed. Finally, it is shown that for a monadic schema (i.e., every method has exactly one parameter), this sufficient condition is also a necessary one.
Yasunori ISHIHARA Kengo MORI Toru FUJIWARA
Detecting the possibility of inference attacks is necessary in order to keep a database secure. Inference attacks mean that a user tries to infer the result of an unauthorized queries to the user. For method schemas, which are a formal model of object-oriented databases, it is known that the security problem against inference attacks is decidable in polynomial time in the size of a given database instance. However, when the database instance or authorization has slightly been updated, it is not desirable to check the entire database again for efficiency. In this paper, we propose several sufficient conditions for update operations to preserve the security. Furthermore, we show that some of the proposed sufficient conditions can be decided much more efficiently than the entire security check. Thus, the sufficient conditions are useful for incremental security checking.