Federated learning (FL) enables distributed collaborative model training without centralizing client data, but its decentralized architecture introduces significant vulnerabilities to backdoor attacks. Malicious clients can embed triggers that compromise the global model’s behavior on specific inputs while maintaining normal performance on clean data. However, existing Byzantine-tolerant defense mechanisms struggle to effectively distinguish malicious updates from benign ones in realistic non-IID environments. This paper presents FedMVC, a two-stage backdoor defense framework for Federated Learning that combines Multi-feature Variational Autoencoders with Clustering to proactively detect and remove malicious client updates before aggregation. The server-side VAE learns compact latent representations of client updates, while clustering applied to the multi-dimensional feature space allows robust discrimination between benign and malicious clusters. We evaluate FedMVC on MNIST, CIFAR-10, and IMDb under both standard backdoor attacks (2 $$\times $$ 2 pixel-block, 10% random-noise) and stronger attacks. Across all settings, FedMVC outperforms existing defenses, substantially reducing the attack success rate while maintaining clean accuracy.

In public key infrastructure, a certificate, issued by a certificate authority (CA), is used to guarantee the connection between a user and her/his public key. In order to improve the efficiency, the concept of implicit certificate protocol is introduced by Girault and Gönther. In the existing implicit certificate protocol, a user must issue a certificate request to the CA for each key pair. However, in certain applications (e.g., IoT, sensor networks, and cryptocurrency), a user (or a device) will have multiple public/private key pairs that are related to the same identity. Therefore, the communication cost will be linearly related to the number of key pairs the user has. Furthermore, the storage cost of a large number of certificates is not an ideal property in practice. In this paper, to address the above issues, we proposed two schemes from the most widely used elliptic curve Qu–Vanstone implicit certificate scheme (ECQV). In our first scheme, called M-ECQV I, an ECQV certificate holder, who obtains an ECQV certificate issued by the certificate authority, can further issue multiple credentials with the same identity as ECQV certificate holder and the corresponding key pairs from the ECQV certificate. In our second scheme, called M-ECQV II, it not only supports the comparable functionality of M-ECQV I, but the verifier can ensure that the credentials are only used by the ECQV certificate holder (i.e., these credential are “self-use”) to be suitable to different scenarios. In addition, the security models are well-defined and the rigorous security proofs are also given. Experimental results show that our schemes not only greatly improve the performance, but also reduce the storage cost.