Search

Natural language processing (NLP) backdoor attacks have become a hidden threat to modern NLP applications. Most of the existing defense methods defend against specific types of backdoor attacks, and they generally fail to defend against invisible backdoors with syntactically correct triggers. This paper proposes NLPSweep, a comprehensive defense scheme that can defend against five common types of backdoor attacks, namely, character, word, sentence, homograph, and learnable textual attacks. Specifically, we propose a framework that can discover an effective defense solution without prior knowledge of the attacks. The defense solution is optimized from the framework and can defend against various attacks while ensuring high accuracy. Finally, we verify the effectiveness of NLPSweep on two pretrained models (BERT and XLNET) on three classic datasets (SST-2, IMDB, and OLID) and compare it with five state-of-the-art defense methods, namely, ONION, Pred, RAP, Fine-pruning, and STRIP. The experimental results demonstrate that NLPSweep has an average model accuracy (ACC) greater than 0.922 and that the average attack success rate (ASR) is only 0.202, outperforming the compared methods. Furthermore, NLPSweep is tested on the real-world Yelp dataset and it can effectively defend against backdoor attacks with the ASR less than 0.07 and the ACC greater than 0.973.1 • Existing defense methods only defend against specific backdoor attacks. • Our defense framework can discover an effective defense solution without prior knowledge of the attacks. • Our defense solution can defend against various attacks while ensuring high accuracy. • Tests on three classic datasets show that our defense scheme outperforms recent work. • Our scheme demonstrates operational success on the real-world Yelp dataset, effectively defending against backdoor attacks. [ABSTRACT FROM AUTHOR]

With the increasing popularity of Internet of Energy (IoE), more and more physical devices connected to IoE depend on the information system for energy control and coordination. Under this condition, how to achieve information flow control in IoE becomes a great challenge. Access control encryption (ACE) is a promising technology to address the problem. However, existing ACE requires a centralized sanitizer, hindering its successful application in IoE. In this paper, we construct a new kind of ACE without sanitizers for IoE. We first construct a basic ACE scheme based on number-theoretic assumptions (i.e., DBDH assumption), and this scheme can control not only what users can read but also what they can write. To resist against the quantum attacks, we further construct a more secure ACE scheme based on learning with errors (LWE). We formally prove that our proposed two ACE schemes are secure under the proposed security definition, and we also evaluate the applicability and the efficiency of them in experiments. [ABSTRACT FROM AUTHOR]

• An integrated privacy-preserving architecture is designed for VSNs. Three scenarios that vehicles request data are considered. • We propose two novel mechanisms to evaluate the reliability of message which are broadcasted by vehicles. • An efficient incentive-punishment mechanism is proposed based on game theory. Besides, blockchain technology and consensus mechanism are employed to protect vehicles' credit information. • Pseudonym and tracking mechanism is proposed for VSNs. In vehicular social networks (VSNs), edge stations or cloud service provider can support the traffic services or location services to vehicles. Moreover, the information sharing among vehicles can assist vehicles to avoid traffic accidents and guarantee driving safety. However, it is easy to be threatened in the communication of vehicle-to-edge station, vehicle-to-cloud service provider and vehicle-to-vehicle in VSNs, which leads to the vehicle's privacy leakage easily. Besides, some malicious users may provide untrustworthy information to mislead others just for its selfishness. Hence, in this paper, we propose an efficient, reliable and privacy-preserving scheme based on blockchain for VSNs. In our scheme, pseudonym mechanism is employed to achieve individual anonymization by concealing the vehicles' identity. Moreover, to encourage vehicles to report trustworthy information, incentive-punishment mechanism is proposed. Meanwhile, we propose multi-factors and single-factor weight-based evaluation mechanism to evaluate the reliability of message. Additionally, Practical Byzantine Fault Tolerance (PBFT) and blockchain are also employed to achieve consensus and store the records respectively, which can prevent malicious entities from manipulating on vehicles' reward scores and credit scores. Finally, we analyze the security of the proposed scheme in terms of external attacks, internal attacks, collusion attacks etc. The relevant experimental results are shown that our scheme is feasible and efficient. [ABSTRACT FROM AUTHOR]

As a special case of public key encryption, identity-based encryption (IBE) takes any public known information as public key for encryption and then decrypts a ciphertext by a well-generated private key from private key generator (PKG). Unlike the traditional IBE using a text-based identity (e-mail, etc.) as public key, in this paper, we aim to design a secure, time-saving and space-saving biometric identity-based encryption (BIBE) regarding the biometric-based identity (face, etc.) as public key. To overcome the challenge introduced by the fuzziness of biometric identities, First, we propose a provable-secure inner-product encryption (IPE) with short ciphertext and show the IPE is indistinguishable against selective identity, adaptive chosen-plaintext attack (IND-sID-CPA). Then, we construct a distance-based encryption (DBE) leveraging the proposed IPE and prove that the DBE captures the same security with the underlying IPE. Furthermore, we optimize the proposed DBE so that it also has short private key. We theoretically analyze the overhead of IPE, DBE, and optimized DBE (ODBE) in terms of time, space, and communication complexities. We also conduct experiments to measure the time and space costs of the proposed ODBE, and experimental results validate its effectiveness and efficiency. [ABSTRACT FROM AUTHOR]

Database as a Service ( DBaaS ) economically enables the customers with limited resources to outsource large-scale databases to the cloud. However, it becomes a very challenging problem that how to execute SQL on outsourced databases while protecting the confidentiality of outsourced databases at the same time. In this paper, we present a novel protocol to enable secure and efficient database outsourcing. First, we propose a new cloud database model by introducing computation service providers (CSPs), which can accommodate the conventional DBaaS model; the CSPs undertake most of the postprocessing and reconstruction burden for database query. Next, we design our database outsourcing protocol SecureDBS by leveraging a secret sharing mechanism and a novel tree-based order-preserving encryption (OPE) mechanism. The proposed protocol supports various SQL queries, including selection, update, insertion, and aggregate queries. Further, we present a mechanism to detect malicious behaviors of the cloud. Finally, we theoretically show that our protocol is correct, secure and efficient, and experimentally demonstrate that the protocol is efficient for practical usage. [ABSTRACT FROM AUTHOR]

The theory of Compressive Sensing (CS) enables the compact storage of image datasets which are exponentially generated today. In this application, the high computational complexity CS reconstruction process is considered to be outsourced to the cloud for its abundant computing and storage resources. Although it is promising, how to protect data privacy and simultaneously maintain management of the image remains challenging. To address the challenge, we propose a novel outsourced image reconstruction and identity authentication service in cloud, which integrates the techniques of signal processing in the CS domain and computation outsourcing. In our system, the image CS samples are outsourced to cloud for reduced storage. For privacy, the scheme ensures the cloud to securely reconstruct image without revealing the underlying content. For management, whether the cloud determines to supply the reconstruction service is depending on the identity authentication result. Theoretical analysis and empirical evaluations show a satisfactory security performance and low computational complexity of the proposed system. Besides, experimental results also confirm the feasibility of identity authentication in the CS domain. [ABSTRACT FROM AUTHOR]

Abstract: Recently Kocarev and Tasev proposed to use Chebyshev polynomials over real numbers to design a public key algorithm by employing the semigroup property. Bergamo et al. pointed out that the public key algorithm based on Chebyshev polynomials working on real numbers is not secure and devised an attack which permits to recover the corresponding plaintext from a given ciphertext. Later Kocarev et al. generalized the Chebyshev polynomials from real number fields to finite fields and finite rings to make the public key algorithm more secure and practical. However, we analyzed the period distribution of the sequences generated by the Chebyshev polynomials over finite fields . When the modulus N is prime, we found this algorithm was also not secure and proposed an attack on this algorithm over finite fields. We then proposed some schemes to improve the security. In this paper, we further analyze in detail the period distribution of the sequences generated by Chebyshev polynomials over the integer ring Z N when N is composite. It turns out that the period distribution is poor if N is not chosen properly and there are many small periods, which are not secure in the sense of cryptology. Based on these findings, we devise an attack on the public key algorithm based on Chebyshev polynomials over the integer ring Z N . We also propose some suggestions to avoid this attack. [Copyright &y& Elsevier]

Machine learning models trained on sensitive real-world data promise improvements to everything from medical screening to disease outbreak discovery. In many application domains, learning participants would benefit from pooling their private datasets, training precise machine learning models on the aggregate data, and sharing the profits of using these models. Considering privacy and security concerns often prevent participants from contributing sensitive data for training, researchers proposed several techniques to achieve data privacy in federated learning systems. However, such techniques are susceptible to causative attacks, whereby malicious participants can inject false training results with the aim of corrupting the well-learned model. To end this, in this paper, we propose a new privacy-preserving federated learning scheme that guarantees the integrity of deep learning processes. Based on the Trusted Execution Environment (TEE), we design a training-integrity protocol for this scheme, in which causative attacks can be detected. Thus, each participant is compelled to execute the privacy-preserving learning algorithm of the scheme correctly. We evaluate the performance of our scheme by prototype implementations. The experimental result shows that the scheme is training-integrity and practical. [ABSTRACT FROM AUTHOR]