Biblio

Recent works have empirically shown that neural network interpretability is susceptible to malicious manipulations. However, existing attacks against Interpretable Deep Learning Systems (IDLSes) all focus on the white-box setting, which is obviously unpractical in real-world scenarios. In this paper, we make the first attempt to attack IDLSes in the decision-based black-box setting. We propose a new framework called Dual Black-box Adversarial Attack (DBAA) which can generate adversarial examples that are misclassified as the target class, yet have very similar interpretations to their benign cases. We conduct comprehensive experiments on different combinations of classifiers and interpreters to illustrate the effectiveness of DBAA. Empirical results show that in all the cases, DBAA achieves high attack success rates and Intersection over Union (IoU) scores.

Nowadays, many data owners choose to outsource their data to public cloud servers while allowing authorized users to retrieve them. To protect data confidentiality from an untrusted cloud, many studies on searchable encryption (SE) are proposed for privacy-preserving search over encrypted data. However, most of the existing SE schemes only focus on the single-owner model. Users need to search one-by-one among data owners to retrieve relevant results even if data are from the same cloud server, which inevitably incurs unnecessary bandwidth and computation cost to users. Thus, how to enable efficient authorized search over multi-owner datasets remains to be fully explored. In this paper, we propose a new privacy-preserving search scheme for the multi-owner and multi-user model. Our proposed scheme has two main advantages: 1) We achieve an attribute-based keyword search for multi-owner model, where users can only search datasets from specific authorized owners. 2) Each data owner can enforce its own fine-grained access policy for users while an authorized user only needs to generate one trapdoor (i.e., encrypted search keyword) to search over multi-owner encrypted data. Through rigorous security analysis and performance evaluation, we demonstrate that our scheme is secure and feasible.

Crowdsensing, driven by the proliferation of sensor-rich mobile devices, has emerged as a promising data sensing and aggregation paradigm. Despite useful, traditional crowdsensing systems typically rely on a centralized third-party platform for data collection and processing, which leads to concerns like single point of failure and lack of operation transparency. Such centralization hinders the wide adoption of crowdsensing by wary participants. We therefore explore an alternative design space of building crowdsensing systems atop the emerging decentralized blockchain technology. While enjoying the benefits brought by the public blockchain, we endeavor to achieve a consolidated set of desirable security properties with a proper choreography of latest techniques and our customized designs. We allow data providers to safely contribute data to the transparent blockchain with the confidentiality guarantee on individual data and differential privacy on the aggregation result. Meanwhile, we ensure the service correctness of data aggregation and sanitization by delicately employing hardware-assisted transparent enclave. Furthermore, we maintain the robustness of our system against faulty data providers that submit invalid data, with a customized zero-knowledge range proof scheme. The experiment results demonstrate the high efficiency of our designs on both mobile client and SGX-enabled server, as well as reasonable on-chain monetary cost of running our task contract on Ethereum.

Multi-Writer Searchable Encryption, also known as public-key encryption with keyword search(PEKS), serves a wide spectrum of data sharing applications. It allows users to search over encrypted data encrypted via different keys. However, most of the existing PEKS schemes are built on classic security assumptions, which are proven to be untenable to overcome the threats of quantum computers. To address the above problem, in this paper, we propose a lattice-based searchable encryption scheme from the learning with errors (LWE) hardness assumption. Specifically, we observe that the keys of each user in a basic scheme are composed of large-sized matrices and basis of the lattice. To reduce the complexity of key management, our scheme is designed to enable users to directly use their identity for data encryption. We present several optimization techniques for implementation to make our design nearly practical. For completeness, we conduct rigorous security, complexity, and parameter analysis on our scheme, and perform comprehensive evaluations at a commodity machine. With a scenario of 100 users, the cost of key generation for each user is 125s, and the cost of searching a document with 1000 keywords is 13.4ms.

Data outsourcing to cloud has been a common IT practice nowadays due to its significant benefits. Meanwhile, security and privacy concerns are critical obstacles to hinder the further adoption of cloud. Although data encryption can mitigate the problem, it reduces the functionality of query processing, e.g., disabling SQL queries. Several schemes have been proposed to enable one-dimensional query on encrypted data, but multi-dimensional range query has not been well addressed. In this paper, we propose a secure and scalable scheme that can support multi-dimensional range queries over encrypted data. The proposed scheme has three salient features: (1) Privacy: the server cannot learn the contents of queries and data records during query processing. (2) Efficiency: we utilize hierarchical cubes to encode multi-dimensional data records and construct a secure tree index on top of such encoding to achieve sublinear query time. (3) Verifiability: our scheme allows users to verify the correctness and completeness of the query results to address server's malicious behaviors. We perform formal security analysis and comprehensive experimental evaluations. The results on real datasets demonstrate that our scheme achieves practical performance while guaranteeing data privacy and result integrity.

Data deduplication [3] is able to effectively identify and eliminate redundant data and only maintain a single copy of files and chunks. Hence, it is widely used in cloud storage systems to save the users' network bandwidth for uploading data. However, the occurrence of deduplication can be easily identified by monitoring and analyzing network traffic, which leads to the risk of user privacy leakage. The attacker can carry out a very dangerous side channel attack, i.e., learn-the-remaining-information (LRI) attack, to reveal users' privacy information by exploiting the side channel of network traffic in deduplication [1]. In the LRI attack, the attacker knows a large part of the target file in the cloud and tries to learn the remaining unknown parts via uploading all possible versions of the file's content. For example, the attacker knows all the contents of the target file X except the sensitive information \texttheta. To learn the sensitive information, the attacker needs to upload m files with all possible values of \texttheta, respectively. If a file Xd with the value \textthetad is deduplicated and other files are not, the attacker knows that the information \texttheta = \textthetad. In the threat model of the LRI attack, we consider a general cloud storage service model that includes two entities, i.e., the user and cloud storage server. The attack is launched by the users who aim to steal the privacy information of other users [1]. The attacker can act as a user via its own account or use multiple accounts to disguise as multiple users. The cloud storage server communicates with the users through Internet. The connections from the clients to the cloud storage server are encrypted by SSL or TLS protocol. Hence, the attacker can monitor and measure the amount of network traffic between the client and server but cannot intercept and analyze the contents of the transmitted data due to the encryption. The attacker can then perform the sophisticated traffic analysis with sufficient computing resources. We propose a simple yet effective scheme, called randomized redundant chunk scheme (RRCS), to significantly mitigate the risk of the LRI attack while maintaining the high bandwidth efficiency of deduplication. The basic idea behind RRCS is to add randomized redundant chunks to mix up the real deduplication states of files used for the LRI attack, which effectively obfuscates the view of the attacker, who attempts to exploit the side channel of network traffic for the LRI attack. RRCS includes three key function modules, range generation (RG), secure bounds setting (SBS), and security-irrelevant redundancy elimination (SRE). When uploading the random-number redundant chunks, RRCS first uses RG to generate a fixed range [0,$łambda$N] ($łambda$ $ε$ (0,1]), in which the number of added redundant chunks is randomly chosen, where N is the total number of chunks in a file and $łambda$ is a system parameter. However, the fixed range may cause a security issue. SBS is used to deal with the bounds of the fixed range to avoid the security issue. There may exist security-irrelevant redundant chunks in RRCS. SRE reduces the security-irrelevant redundant chunks to improve the deduplication efficiency. The design details are presented in our technical report [5]. Our security analysis demonstrates RRCS can significantly reduce the risk of the LRI attack [5]. We examine the performance of RRCS using three real-world trace-based datasets, i.e., Fslhomes [2], MacOS [2], and Onefull [4], and compare RRCS with the randomized threshold scheme (RTS) [1]. Our experimental results show that source-based deduplication eliminates 100% data redundancy which however has no security guarantee. File-level (chunk-level) RTS only eliminates 8.1% – 16.8% (9.8% – 20.3%) redundancy, due to only eliminating the redundancy of the files (chunks) that have many copies. RRCS with $łambda$ = 0.5 eliminates 76.1% – 78.0% redundancy and RRCS with $łambda$ = 1 eliminates 47.9% – 53.6% redundancy.

Modern distributed key-value stores are offering superior performance, incremental scalability, and fine availability for data-intensive computing and cloud-based applications. Among those distributed data stores, the designs that ensure the confidentiality of sensitive data, however, have not been fully explored yet. In this paper, we focus on designing and implementing an encrypted, distributed, and searchable key-value store. It achieves strong protection on data privacy while preserving all the above prominent features of plaintext systems. We first design a secure data partition algorithm that distributes encrypted data evenly across a cluster of nodes. Based on this algorithm, we propose a secure transformation layer that supports multiple data models in a privacy-preserving way, and implement two basic APIs for the proposed encrypted key-value store. To enable secure search queries for secondary attributes of data, we leverage searchable symmetric encryption to design the encrypted secondary indexes which consider security, efficiency, and data locality simultaneously, and further enable secure query processing in parallel. For completeness, we present formal security analysis to demonstrate the strong security strength of the proposed designs. We implement the system prototype and deploy it to a cluster at Microsoft Azure. Comprehensive performance evaluation is conducted in terms of Put/Get throughput, Put/Get latency under different workloads, system scaling cost, and secure query performance. The comparison with Redis shows that our prototype can function in a practical manner.

The image and multimedia data produced by individuals and enterprises is increasing every day. Motivated by the advances in cloud computing, there is a growing need to outsource such computational intensive image feature detection tasks to cloud for its economic computing resources and on-demand ubiquitous access. However, the concerns over the effective protection of private image and multimedia data when outsourcing it to cloud platform become the major barrier that impedes the further implementation of cloud computing techniques over massive amount of image and multimedia data. To address this fundamental challenge, we study the state-of-the-art image feature detection algorithms and focus on Scalar Invariant Feature Transform (SIFT), which is one of the most important local feature detection algorithms and has been broadly employed in different areas, including object recognition, image matching, robotic mapping, and so on. We analyze and model the privacy requirements in outsourcing SIFT computation and propose Secure Scalar Invariant Feature Transform (SecSIFT), a high-performance privacy-preserving SIFT feature detection system. In contrast to previous works, the proposed design is not restricted by the efficiency limitations of current homomorphic encryption scheme. In our design, we decompose and distribute the computation procedures of the original SIFT algorithm to a set of independent, co-operative cloud servers and keep the outsourced computation procedures as simple as possible to avoid utilizing a computationally expensive homomorphic encryption scheme. The proposed SecSIFT enables implementation with practical computation and communication complexity. Extensive experimental results demonstrate that SecSIFT performs comparably to original SIFT on image benchmarks while capable of preserving the privacy in an efficient way.