Qianqian Su

The Transformer model has emerged as a prominent machine learning tool within the field of natural language processing. Nevertheless, running the Transformer model on resource-constrained devices presents a notable challenge that needs to be addressed. Although outsourcing services can significantly reduce the computational overhead associated with using the model, it also incurs privacy risks to the provider’s proprietary model and the client’s sensitive data. In this paper, we propose an efficient privacy-preserving Transformer inference framework (PPTIF) for language translation tasks based on three-party replicated secret-sharing techniques. PPTIF offers a secure approach for users to leverage Transformer-based applications, such as language translation, while maintaining the confidentiality of their original input and inference results, thereby preventing any disclosure to the cloud server. Meanwhile, PPTIF ensures robust protection for the model parameters, guaranteeing their integrity and confidentiality. In PPTIF, we design a series of interaction protocols to implement the secure computation of Transformer components, namely secure Encoder and secure Decoder. To improve the efficiency of PPTIF, we optimize the computation of the Scaled Dot-Product Attention (Transformer’s core operation) under secret sharing, effectively reducing its computation and communication overhead. Compared with Privformer, the optimized Masked Multi-Head Attention achieves about 1.7× lower runtime and 2.3× lower communication. In total, PPTIF achieve about 1.3× lower runtime and 1.2× lower communication. The effectiveness and security of PPTIF have been rigorously evaluated through comprehensive theoretical analysis and experimental validation.

Online neural network prediction services provide a convenient way for users to use neural network models. In this paradigm, the model owner pre-deploys the trained model to the cloud server, and the user just uploads the data to get the prediction results. However, uploading models and data to an unreliable cloud also brings up serious privacy concerns, such as model theft and sensitive information leakage. In this paper, we propose a securely outsourced framework (ESANNP for short) for securely online neural network prediction. The framework is built utilizing the additive homomorphic encryption method under a dual-server non-collusion architecture. First, we design a series of high-precision fixed-point secure computing protocols. These protocols are crucial for securely carrying out complex nonlinear activation functions like Sigmoid, Tanh, Softmax, etc. Second, we construct the framework ESANNP based on the designed protocols. The ESANNP can implement many types of neural networks for online prediction while guaranteeing the security of data and model data. In addition, to support the needs of different security levels, we have set up a mechanism in ESANNP to allow users to choose their own security parameters according to their needs and complete data privacy protection. We demonstrate that the proposed ESANNP can accomplish secure online neural network prediction without revealing the model and the user data’s privacy. Finally, we also evaluate the efficiency and accuracy of the proposed protocols through extensive experiments. We apply the ESANNP to a fully connected neural network containing 238,510 weights and using Sigmoid as the activation function. It takes about 7 minutes for secure prediction, and the maximum error of the prediction results does not exceed 10−7.

Smart grid is gaining more and more attention as one of the typical applications of Internet of Things. However, in a distributed environment, how to guarantee the privacy of users in electricity trading while ensuring the efficiency of the transactions is one of the urgent issues to be solved. In this article, we propose a distributed attribute-based signature (DABS) scheme for distributed electricity trading, which can support users' free choice of trade objects without revealing their real identities. We construct a signature generation and verification method by taking advantage of the open and hard-to-tamper properties of blockchain to achieve signature verifiability independent of dynamic changes in attributes.To improve the update efficiency, we propose an improved scheme that enables the update complexity to be reduced from $O(n)$ to $O(\log n)$ , where $n$ is the number of users. Finally, performance analysis and simulation experiments demonstrate the security and practicality of the DABS.

Traditional identity authentication solutions mostly rely on a trusted central entity, so they cannot handle single points of failure well. In addition, most of these traditional schemes need to store a large amount of identity authentication or public key information, which makes the schemes difficult to expand and use in distributed situations. In addition, the user prefers to protect the privacy of their information during the identity verification process. Due to the open and decentralized nature of the blockchain, the existing identity verification schemes are difficult to apply well in the blockchain. To solve this problem, in this article, we propose a privacy protection identity authentication scheme based on the blockchain. The user independently generates multiple-identity information, and these identities can be used to apply for an identity certificate. Authorities use the ECDSA signature algorithm and the RSA encryption algorithm to complete the distribution of the identity certificate based on the identity information and complete the registration of identity authentication through the smart contract on the blockchain. On the one hand, it can realize the protection of real identity information; on the other hand, it can avoid the storage overhead caused by the need to store a large number of certificates or key pairs. Due to the use of the blockchain, there is no single point of failure in the authentication process, and it can be applied to distributed scenarios. The security and performance analysis show that the proposed scheme can meet security requirements and is feasible.