Secure and Privacy-Aware Infrastructure Log Synthesis Using Federated Generative Adversarial Networks

This research introduces a privacy-preserving system that enables entities to share network logs for increased transparency and collaboration in threat intelligence. Cybersecurity analytics is very dependent on network logs for the detection of threats and the analysis of system behavior. Unfortunately, due to the fact that the privacy and security risks are doubled, many organizations either do not have the capability to share network logs or are just unwilling to do so. To resolve this problem, the paper suggests employing Federated Generative Adversarial Networks (FedGAN) in order to create a worldwide generative model that can generate plausible synthetic network logs without the need for any data exposure. In the proposed system, the multiple clients are able to jointly train a common GAN model with each node simply sending the updates to the model which ensures that the privacy of the data is preserved. The GAN generator gets trained to produce network traffic that retains the statistical and behavioral properties of the original traffic, while the GAN discriminator identifies whether the logs are synthetic or real. The experiments that have been performed with standard network datasets show the production of synthetic logs that are very close to real ones in terms of various metrics such as event occurrence, temporal correlations, and anomalous distribution. To conclude, when synthetic logs are injected into an intrusion detection and anomaly analysis system, models trained thereupon show results nearly as good as those of real-datatrained models, thus proving their value.