From the course: AI for Telecom: Network Optimization and Security in 5G/Edge Systems
Join today to access over 25,500 courses taught by industry experts.
Solution: Flow-based anomaly detection
From the course: AI for Telecom: Network Optimization and Security in 5G/Edge Systems
Solution: Flow-based anomaly detection
(bright music) - [Instructor] Here is the solution. An AI powered one, which is flow-based anomaly detection using a self-supervised learning model. That can learn normal traffic behavior purely from the flow data such as IP addresses, flow duration, and package sizes. Let me give you more details. The solution components could be a data collection where you can use NetFlow/IPFIX records and 5G GTP-U metadata. You can also have feature engineering where you can extract temporal and statistical features such as entropy of flow direction and RTT variation. In the model, you can use transformer-based autoencoder to reconstruct the traffic patterns. For the inference, you can reconstruct error which can exceeds the threshold, so it can flag the flow as anomalous. For the outcome, you'll be amazed to see. You can identify the CPEs acting as proxy exit nodes. You can also detect volumetric exfiltration from compromised IoT clusters. And lastly, you'll be able to adapt to evolving traffic…
Download courses and learn on the go
Watch courses on your mobile device without an internet connection. Download courses using your iOS or Android LinkedIn Learning app.