CATO

End-to-end ML traffic analysis pipeline optimizer

Overview

CATO framework for jointly optimizing predictive performance and systems costs of ML-based traffic analysis pipelines.

Features

  • Multi-objective Bayesian optimization
  • Pareto-optimal configuration identification
  • Automated pipeline compilation
  • Feature and window size optimization
  • Real-time deployment support

Key Capabilities

  • Balances accuracy and inference latency
  • Reduces computational overhead
  • Optimizes feature selection
  • Minimizes data collection requirements

Applications

  • IoT device classification
  • Real-time traffic analysis
  • Network security monitoring
  • QoS management

Citation

@article{wan2025cato,
  title={CATO: End-to-end Optimization of ML Traffic Analysis Pipelines},
  author={Wan, Gerry and Liu, Shinan and Bronzino, Francesco and Feamster, Nick and Durumeric, Zakir},
  journal={USENIX NSDI},
  year={2025}
}