Backdoor attacks pose a critical threat to neural network models, allowing attackers to implant a backdoor during the training phase by manipulating a small portion of the training data. In security-sensitive applications such as voice interaction for autonomous driving, the presence of backdoor attacks introduces substantial security risks. This study focuses on implementing backdoor defense measures for speech recognition models in run-time, taking into account the characteristics of audio signals. We propose SpeechGuard, the first online backdoor defense pipeline designed to identify and purify poisoned audio samples. Specifically, we improve STRIP method to perform adaptive perturbation injection to detect and filter poisoned samples, named as S-STRIP. More importantly, we further consider the purification of poisoned samples. We utilize time-frequency (T-F) masking to suppress the expression of trigger signals and autonomously generate masks based on an autoencoder. The two-stage processing prevents the backdoor in the model from being triggered, and even input speech carrying triggers can be accurately predicted. Extensive experimental demonstrate that SpeechGuard can accurately filter out poisoned samples. Through purification, it can significantly mitigate the backdoor threat while maintaining a certain prediction accuracy.

Federated Learning (FL) solves the privacy leakage problem of cross-domain training, but is vulnerable to Byzantine attacks, that is, the adversary uses poisoned data or poisoned local models that affect the convergence of the global model. However, most existing Byzantine-robust methods either do not converge, or are only suitable for honest majority settings and cannot cope with well-designed malicious gradients. To this end, we propose a Attention-based Byzantine-robust Federated Learning with Vector Filter (AFL), which is 1.1× to 4.2× more resistant to attacks on different datasets and models than other Byzantine defense schemes. Specifically, we bootstrap trust with a clean root dataset, and propose a new scaled dot-product attention-based aggregation method, which accelerates global model convergence while maximally preserving benign gradients with uneven quality. Additionally, we add a vector filter as the first line of defense against malicious gradients. We conduct extensive evaluations using the MNIST, FMNIST and CIFAR-10 datasets, and the experimental results show that the AFL scheme improves the average test accuracy, works well even with 80% malicious Byzantine adversaries. The AFL scheme can deal with state-of-the-art model poisoning attacks, and has high robustness and effectiveness.

In the scenario of the Internet of Things (IoT) co-located with the cloud, many applications, such as face recognition, traffic monitoring, and medical diagnosis, usually outsource a large amount of generated image data to cloud servers (CSs) to reduce the burden of local storage. Many secure encrypted image retrieval schemes have been proposed to protect data privacy. However, existing work incurs storage and communication burdens, lacks verifiability of query results and has potential forward security threats. To solve these issues, we propose the VerFHS framework in this article, which can satisfy Verifiability, Feedback, and High-Security. Specifically, we first present an extended secure$k$-NN algorithm to protect indexes, cleverly design ciphertext inner product for similarity comparison, and use the reward mechanism of blockchain to build a monitoring and feedback mechanism for CSs. Then we demonstrate an enhanced VerFHS scheme in the dynamic setting (VerFHSD) that uses a permutation matrix to process image encryption against adaptive attacks during dynamic updates. VerFHSD prevents CSs from making search queries over newly added images via previous tokens, thereby achieving forward security. The formal security analysis shows that our schemes protect the privacy of images, indexes and query tokens, and forward security. And extensive experiments using the real-world data set demonstrate that our scheme not only has the highest search accuracy all the time, but also achieves efficient queries at the millisecond level, when compared with other advanced image retrieval schemes.

Message-Locked Encryption (MLE) is a useful tool to enable deduplication over encrypted data in cloud storage. It can significantly improve the cloud service quality by eliminating redundancy to save storage resources, and hence user cost, and also providing defense against different types of attacks, such as duplicate faking attack and brute-force attack. A typical MLE scheme only focuses on deduplication. On the other hand, supporting search operations on stored content is another essential requirement for cloud storage. In this article, we present a message-locked searchable encryption (MLSE) scheme in a dual-server setting, which achieves simultaneously the desirable features of supporting deduplication and enabling users to perform search operations over encrypted data. In addition, it supports both multi-keyword and negative keyword searches. We formulate the security notions of MLSE and prove our scheme satisfies all the security requirements. Moreover, we provide an interesting extension of our construction to support Proof of Storage (PoS). Compared with the existing solutions, MLSE achieves better functionalities and efficiency, and hence enables more versatile and efficient cloud storage service.

The existence of adversarial examples and the easiness with which they can be generated raise several security concerns with regard to deep learning systems, pushing researchers to develop suitable defence mechanisms. The use of networks adopting error-correcting output codes (ECOC) has recently been proposed to counter the creation of adversarial examples in a white-box setting. In this paper, we carry out an in-depth investigation of the adversarial robustness achieved by the ECOC approach. We do so by proposing a new adversarial attack specifically designed for multilabel classification architectures, like the ECOC-based one, and by applying two existing attacks. In contrast to previous findings, our analysis reveals that ECOC-based networks can be attacked quite easily by introducing a small adversarial perturbation. Moreover, the adversarial examples can be generated in such a way to achieve high probabilities for the predicted target class, hence making it difficult to use the prediction confidence to detect them. Our findings are proven by means of experimental results obtained on MNIST, CIFAR-10, and GTSRB classification tasks.

The vehicular reputation management scheme based on blockchain has been studied with the rapidly development of blockchain technology. Existing works mainly focus on provisioning the security of the reputation management scheme by using the blockchain technology. However, the broadcast latency of blocks and the throughput capacity of the blockchain also have significant impacts on the performance of the reputation management scheme. This paper proposes a hierarchical blockchain based reputation management scheme in vehicular networks. Our hierarchical blockchain architecture assures that each vehicle can obtain the latest reputation value of any neighboring vehicle in a short time. Moreover, our designed blockchain structure aims at reducing the confirmation time of reputation values and increasing the capacity of blockchain. We finally conduct simulation experiments to compare our proposed scheme with the existing schemes.

The physical layer security of OFDM communication systems is getting more and more attention. Protecting the channel from eavesdroppers has received some attention. Artificial noise is generally considered to utilize the null space of the intended receiver's channel to interfere with the eavesdropper's channel, but it may not be able to defend against passive eavesdroppers. In this paper, a novel anti-eavesdropping OFDM system is proposed by using artificial pilot noise. Pilots play an important role in demodulating OFDM signal. In our proposed scheme, an auxiliary device called Helper would generate noise signals that overlap on the pilots sent by the user terminal. With the knowledge of the signals sent by the Helper, the receiver can decode the user terminal signal but eavesdroppers cannot. This would prevent eavesdroppers from overhearing the information transmitted between the user terminal and the receiver, so as to provide a secure communication for the user terminal. Finally, we use wireless open access research platform (WARP) v3 to test the feasibility and security of the scheme in practice. Our experimental results show that the proposed scheme can provide a secure communication in the environments with passive eavesdroppers.

Bundle authentication, which ensures the authenticity and integrity of bundles, is critical in space Delay/disruption-Tolerant Networks (DTNs). When bundle fragment services are needed, the previous solutions directly using digital signatures suffer from heavy computational costs and bandwidth overheads. This paper addresses the issue of fragment authentication for Bundle Protocol by exploiting erasure codes and the batch transmission characteristic of DTNs. Erasure codes are adopted to allow all the fragments of a bundle to equally share only one signature, to tolerate high delays as well as unexpected loss of connectivity. Following this generic idea, we present two approaches, both of which are effective in filtering inauthentic fragments as early as possible. The first one takes a surprisingly low bandwidth overhead, while it makes all received fragments of a bundle to be removed when there is an inauthentic one, because of its failure in locating the inauthentic fragments. Considering this defect, we present an improved scheme which is able to detect inauthentic fragments thanks to a special hash chain and then only remove these inauthentic ones. The performance simulation demonstrates that both our schemes significantly reduce bandwidth overheads and computational costs as compared to the prior works.

ABSTRACT Consider a signed file that cannot be verified because of some fraction loss or noise, which might not affect the business of users. It would be desirable for the verifier to locate the damaged fractions so that the authenticity of other components of the file can be ensured. Thereafter, the verifier can decide whether or not to accept it. To address this issue, in this paper, we present a loss‐tolerant authentication scheme, which allows the verifier to locate the damaged elements in the signed file stream. Our idea is to allow the signed file to be fragmented, and each fragment is appended a short authentication tag, which enables the verifier to find the modified fragments in the signed file and verify the signature even if some fragments are lost. The proposed scheme adopts a combination of Reed–Solomon error correction and erasure codes, aiming to allow verifiers to locate the modified fractions and thereafter reconstruct the signature when sufficient original fractions of the file are obtained. Our scheme offers many potential applications such as loss‐tolerant transmission and storage. Copyright © 2013 John Wiley & Sons, Ltd.

ABSTRACT Public key cryptography is an important technique to resolve the security issues in computer communication networks. In some scenarios, a public key cryptosystem with double trapdoor decryption mechanism is desired. Previous known public key cryptosystems with double trapdoor decryption mechanism achieve no standard semantic security goals against adaptive chosen ciphertext attacks and were shown vulnerable to some attacks. In this paper, two new problems, the generalized RSA problem and the computational generalized RSA problem, are introduced, and the equivalence of the standard RSA problem and the computational generalized RSA problem is proven. Accordingly, on the basis of the computational generalized RSA problem, a public key cryptosystem with double trapdoor decryption mechanism, called GenRSA, is presented. We prove the GenRSA's indistinguishability security against adaptive chosen ciphertext attack under the computational generalized RSA intractability assumption and in the random oracle model. Copyright © 2013 John Wiley & Sons, Ltd.

To ensure the authenticity, integrity, and confidentiality of bundles, the in-transit Protocol Data Units of bundle protocol (BP) in space delay/disruption tolerant networks (DTNs), the Consultative Committee for Space Data Systems bundle security protocol (BSP) specification suggests four IPsec style security headers to provide four aspects of security services. However, this specification leaves key management as an open problem. Aiming to address the key establishment issue for BP, in this paper, we utilize a time-evolving topology model and two-channel cryptography to design efficient and noninteractive key exchange protocol. A time-evolving model is used to formally model the periodic and predetermined behavior patterns of space DTNs, and therefore, a node can schedule when and to whom it should send its public key. Meanwhile, the application of two-channel cryptography enables DTN nodes to exchange their public keys or revocation status information, with authentication assurance and in a noninteractive manner. The proposed scheme helps to establish a secure context to support for BSP, tolerating high delays, and unexpected loss of connectivity of space DTNs.

Secure group communication with efficient self‐organising key agreement protocol and small broadcast ciphertext is essential to many collaborative and distributed applications in mobile ad‐hoc networks (MANETs). In this study, the authors propose a Chinese Remainder Theorem‐based secure group communication scheme which is able to provide confidentiality service and non‐repudiation service simultaneously. In the proposed scheme, all group members contribute their own public keys to negotiate a shared encryption public key, which corresponds to all different decryption keys. By using the shared public key and the respective secret key, confidentiality and non‐repudiation can be obtained, respectively. Both are essential to secure group communication in MANETs. Compared with the exiting signcryption schemes for multiple recipients, the proposed self‐organising scheme has much smaller ciphertext that does not increase with the size of the group. Accordingly, it is quite suitable to secure group communication in self‐organising, distributed and resource‐constrained MANETs.

ABSTRACT A certificateless public key cryptosystem can make a special contribution to building key distribution and management architecture in resource‐constrained mobile ad hoc networks (MANETs) because it has no separate certificate and no complex certificate management problems. In this paper, we present a virtual private key generator (VPKG)‐based escrow‐free certificateless public key cryptosystem as a novel combination of certificateless and threshold cryptography. Using secret sharing, we build a VPKG whose members collaboratively calculate the partial private key and send it to the user via public channels. The private key of a user is generated jointly by the VPKG and the user. Each of them has “half” of the secret information about the private key of the user. In addition, binding a user's public key with its identity and partial private key, respectively, raises our schemes to the same trust level as is enjoyed in a traditional public key infrastructure. We also show that the proposed scheme is secure against public key replacement attacks and passive attacks. Copyright © 2012 John Wiley & Sons, Ltd.

Since its introduction, broadcast encryption has attracted many useful applications. One major problem faced by such broadcast systems is "piracy." An effective approach to piracy prevention in such systems is traitor tracing. In this paper, a new traitor tracing scheme is proposed by using the efficient and computationally inexpensive public key cryptosystem NTRU. The proposed scheme is the first concrete construction of NTRU-based public-key traitor tracing and has the advantages of extremely efficient encryption and decryption, fast and easy key creation, low memory requirements and revocation property etc. Moreover, this novel work contains other desirable features, such as Collusion-resistant, which the previous traitor tracing schemes did not offer. In addition, with its complexity only O(log2n), the tracing algorithm of this system is more efficient than that of the previous ones.