CascadeNet: Generating Network Traffic with High-Fidelity Temporal Patterns
summary
CascadeNet generates traces by converting each flow into an aggregated per-flow time series and then using CascadeGAN hierarchical generators (MLP → LSTM → MLP) to produce flow tuple, time series, and record fields conditioned on sliding windows. It includes three timestamp-recovery methods with the ML-based timestamp recovery as default. Evaluations on four IPv4 traces show 41–76% temporal improvement, 7.3×–25× faster training and 9–62× faster generation, and 99.9% novelty versus 1.7–56.5% memorization by a competitive baseline.
AI-written summary · about
chapters (14)
- 0:00CascadeNet introduction
- 0:43Synthetic traces motivation
- 2:07ML generator limits
- 3:34Temporal pattern insight
- 4:15CascadeNet workflow
- 5:22Cascade GAN design
- 7:32Post-processing challenges
- 8:56Evaluation datasets
- 9:40Temporal fidelity results
- 10:47Statistical fidelity tasks
- 11:51Burst analysis prediction
- 12:33Anomaly detection
- 12:53Diversity scalability
- 13:57Summary
related talks
40mHow to Map Networked Devices Using Graph-Based AI to Preempt Cyber Threats
16mNetwork Flow Data in the Cloud
16mResolving Packets from Counters: Enabling Multi-scale Network Traffic Super Resolution...
16mScalable Tail Latency Estimation for Data Center Networks
29mFinding the Needle in a Haystack: Tenant-Level Network...
51mAdvanced approaches to traffic inspection & network diagnosis w/
46mSelf-Programming Networks: Network Clock Synchronization, Event Playback and Real-time Action
14mBrain-on-Switch: Towards Advanced Intelligent Network Data Plane via NN-Driven Traffic...