b'Advanced MachineNovel machine learning algorithm increases the detection Learning-based Fifth- accuracy of malicious traffic on encrypted 5G networks.Generation Network AttackT he goal of this project was to prove the principle that network attacks can be detected on a fifth-generation (5G) network using machine learning. Detection System To achieve this goal, we created multiple novel data sets that contained network traffic data alongside multiple types of malicious traffic. One of these data sets was created on a traditional network to simulate modern attacks, which is missing in most academic data sets. The other data set was created on a 5G network and includes novel attacks on the 5G core. In conjunction with these efforts, we tested the common academic approaches of network traffic analysis on the most PROJECT NUMBER:used academic network packet data sets. We then applied multiple techniques from 21A1050-026 other fields of machine learning and artificial intelligence, including computer vision and natural language processing, to see if any of those methods could be applied TOTAL APPROVED AMOUNT:in our setting. After testing many different structures of models, we determined $725,000 over 2 years that variational autoencoders provided the best results. During this analysis, we PRINCIPAL INVESTIGATOR:discovered the imbalanced nature of the data we and others have collected. Many Jared Wadsworth of the common metrics are skewed toward lab environments and do not perform as well on live networks. To combat this issue, we created a generative adversarial CO-INVESTIGATORS: network capable of mimicking both normal and malicious traffic. This allowed us Kurt Durr, INL to train our models on synthetic data in addition to real data, which allowed our Shad Staples, INL models to train longer without biasing the results.Architecture for detecting attacks at the edge of the 5G network via a machine learning / artificial intelligence algorithm.98'