CCLab: Adversarial Testing of Learning- and Non-Learning-Based Congestion Controllers
Researchers have developed CCLab, a new framework designed to test the robustness of network congestion controllers, including both learning-based and traditional algorithms. The framework uses a reinforcement learning agent to introduce adversarial perturbations to input signals or network conditions. Findings indicate that while both types of controllers degrade under attack, learning-based methods generally show greater resilience than human-designed ones. The adversarial traces generated by CCLab can also be used to train more robust congestion controllers. AI
IMPACT Introduces a novel testing framework that could lead to more resilient AI-driven network management systems.