Large‐scale data training is vital to the generalization performance of deep learning (DL) models. However, collecting data directly is associated with increased risk of privacy disclosure, particularly in special fields such as healthcare, finance, and genomics. To protect training data privacy, collaborative deep learning (CDL) has been proposed to enable joint training from multiple data owners while providing reliable privacy guarantee. However, recent studies have shown that CDL is vulnerable to several attacks that could reveal sensitive information about the original training data. One of the most powerful attacks benefits from the leakage from gradient sharing during collaborative training process. In this study, we present a new CDL framework, $\mathsf{PrivateDL}$, to effectively protect private training data against leakage from gradient sharing. Unlike conventional training process that trains on private data directly, $\mathsf{PrivateDL}$ allows effective transfer of relational knowledge from sensitive data to public data in a privacy‐preserving way, and enables participants to jointly learn local models based on the public data with noise‐preserving labels. This way, $\mathsf{PrivateDL}$ establishes a privacy gap between the local models and the private datasets, thereby ensuring privacy against the attacks launched to the local models through gradient sharing. Moreover, we propose a new algorithm called Distributed Aggregation Stochastic Gradient Descent, which is designed to improve the efficiency and accuracy of CDL, especially in the asynchronous training mode. Experimental results demonstrate that $\mathsf{PrivateDL}$ preserves data privacy with reasonable performance overhead.

As an extremely significant cryptographic primitive, certificateless signature (CLS) schemes can provide message authentication with no use of traditional digital certificates. High efficiency and provable security without random oracle are challenges in designing a CLS scheme. Recently, Karati, Islam and Karuppiah proposed an efficient pairing-based CLS scheme with no use of map-to-point (MTP) hash function and random oracle model to provide data authenticity in Industrial Internet of Things (IIOT) systems. The security proof was given under several hardness assumptions. However, we notice that both public key replacement attack and known message attack are existing in Karati et al.’s scheme. Any adversary without knowledge of signer’s private key is capable of forging valid signatures. This lead to several serious consequences. For example, anybody can sign IIoT data on behalf of IIoT data owner without being detected.

Multi-party private set intersection (PSI) enables multiple parties to compute the common items of private sets without disclosing any other information beyond the result; thus, it has gained significant importance in various distributed computation scenarios. As the number of participants increases, the performance of multi-party PSI protocols is significantly influenced, primarily by the number of interaction rounds needed. In this study, we propose two novel multi-party PSI protocols: the first one is 1MPSI, and the second is 2MPSI. The 1MPSI appears as a wheel structure where the parties need only one round of interactive communication online. 1MPSI is based on the Ring version of Oblivious Linear-function Evaluation (OLE). Benefiting from the wheel structure, 1MPSI supports parallel computation and achieves competitive efficiency between the leader and the other participants (OLE receivers) after input-independent precomputation. The 2MPSI adopts a dual-core star structure and introduces Oblivious Key-Value Store (OKVS), which results in better performance when handling larger set sizes and more participants. Our protocols are designed with simplicity and ease of implementation in mind. Experimental evaluations demonstrate the superiority of our protocols over current open-source multi-party PSI protocols as the set size increases from 2¹² to 2²⁰ when involving 10 and 16 parties. In a test with 16 parties each inputting 2²⁰ elements, 2MPSI achieves a runtime of only 65 seconds.