A Covert Communication Method Based on Gradient Model
Wang Degang, Yi Sun, Zhou Chuan-xin · 2021 IEEE 6th International Conference on Signal and Image Processing (ICSIP) · 2021
Pdf, word, etc. have always been used for malicious code transmission and data leakage, and academia have performed a lot of research on the detection of these files. However, with the evolution of big data, artificial intelligence and other new technologies, federated learning, as a new generation of distributed machine learning training method, is widely used in cross domain joint training by exchanging model parameters rather than raw data. This makes it possible to transmit malicious code and disclose sensitive data in federated learning. Therefore, this paper innovatively proposes a covert communication method based on gradient model, which uses the least significant bit steganography algorithm to embed data during model parameters exchanges between clients and central server. Experiments demonstrate that the proposed method has nearly no influence on federated learning global model, apart from strong robustness and large covert channel capacity.