As the proportion of encrypted traffic in network communications continues to increase, encryption technologies are widely used to protect user privacy and data security. Meanwhile, this also makes it more covert for hackers to spread malware, steal sensitive information, or conduct other harmful behaviors in the network. How to effectively detect encrypted malicious traffic in communications while protecting user privacy has become a key task to be addressed in the field of network security. To address this challenge, this paper proposes a detection method for malware communication traffic via encrypted traffic analysis. It extracts contextual correlations and temporal features from raw data traffic without decrypting the encrypted data using Session-Transformer. The method uses Deep Neural Networks as the classifier to detect and classify encrypted malicious traffic. The experimental results show that our method has the best performance in accuracy, precision, recall, and F1-score on the DataCon2020 encrypted malicious traffic dataset and the CIC-AndMal-2017 dataset. In particular, the recall on the DataCon2020 dataset reaches 98.34%, and the precision on the CIC-AndMal-2017 dataset achieves 93.54%.

Secure inference is the main technology to avoid privacy leakage when reasoning with machine learning models in multiparty settings for industrial Internet of Things (IoT). Existing solutions based on secure multiparty computation have been extensively explored in both academic and industrial fields, but these solutions are deployed in single user single model provider setting. When multiparty collaborate to handle tasks in multiuser multimodel provider setting, these solutions are no longer applicable as the mechanism of the linear computation protocol restricting the number of parties increases. In addition, these existing schemes need to conduct multiple checks during the inference of malicious models, which results in additional communication overhead. To solve these issues, we propose a multiparty secure inference scheme by using an improved multiplication protocol for achieving matrix multiplication, which achieves efficient linear calculation and supports adaptive parties. We also design a new check protocol to inspect calculation results of all layers with just one execution, which achieves lower communication and calculation overheads. Formal security analysis proves that our scheme achieves malicious security in the honest-majority setting, and extensive experiments demonstrate that our scheme reduces the communication costs by about 30% in large neural networks when compare with previous solutions.

To solve the data silos issue in distributed machine learning with privacy leakage, privacy-preserving federated learning (PPFL) has been extensively explored in both academic and industrial fields. However, the existing PPFL solutions still suffer from high computation and communication overheads, which result in excessive consumption of communication bandwidth and slow down the training process of FL. To address these issues, we propose a secure and communication-efficient FL scheme using improved compressed sensing and CKKS homomorphic encryption. Specifically, we implement a lossy compression of the model by using discrete cosine transform, then use CKKS homomorphic encryption to encrypt the data transmitted between clients and center server due to its high efficiency and support for batch encryption. Formal security analysis proves that our scheme is secure against indistinguishability under chosen plaintext attack and extensive experiments demonstrate that our scheme achieves a high accuracy at 0.05% compression rate.

With the rapid development of geographic location technology and the explosive growth of data, a large amount of spatial data is outsourced to the cloud server for reducing the local high storage and computing burdens, but at the same time causes security issues. Thus, extensive privacy-preserving spatial data query schemes have been proposed. Most of the existing schemes use Asymmetric Scalar-Product-Preserving Encryption (ASPE) to encrypt data, but ASPE has proven to be insecure against known plaintext attack. And the existing schemes require users to provide more information about query range and thus generate a large amount of ciphertexts, which causes high storage and computational burdens. To solve these issues, based on enhanced ASPE designed in our conference version, we first propose a basic Privacy-preserving Spatial Data Query (PSDQ) scheme by using a new unified index structure, which only requires users to provide less information about query range. Then, we propose an enhanced PSDQ scheme (PSDQ$^+$) by using Geohash-based$R$-tree structure (called$GR$-tree) and efficient pruning strategy, which greatly reduces the query time. Formal security analysis proves that our schemes achieve Indistinguishability under Chosen Plaintext Attack (IND-CPA), and extensive experiments demonstrate that our schemes are efficient in practice.