Anomaly-Aware Compiler Optimization: Predictive Guardrails for ML Systems

Building on Hou et al.'s work on MOSFET failure prediction and Lebaku et al.'s anomaly detection in autonomous vehicles, this idea proposes embedding lightweight anomaly-detection models within the compiler itself. Current ML compilers (e.g., TVM, MLGO) optimize for speed/accuracy but ignore silent failure modes like sensor drifts or thermal degradation. By training models to flag code patterns linked to anomalies (e.g., LSTM layers prone to instability under certain optimizations), the compiler could proactively disable risky optimizations (e.g., aggressive operator fusion) and substitute safer alternatives. Unlike BladeDISC’s focus on dynamic shapes, this addresses runtime behavioral anomalies—a gap in existing work. This could prevent crashes in safety-critical ML deployments (e.g., FCHEVs in Song et al.) by trading marginal performance for robustness.

References:

1. Precursor Prediction and Early Warning of Power MOSFET Failure Using Machine Learning With Model Uncertainty Considered. Yuluo Hou, Chang Lu, Waseem Abbas, Mesfin Seid Ibrahim, Muhammad Waseem, Hiu Hung Lee, K. Loo (2024). IEEE Journal of Emerging and Selected Topics in Power Electronics.
2. Cybersecurity-Focused Anomaly Detection in Connected Autonomous Vehicles Using Machine Learning. Prathyush Kumar Reddy Lebaku, Lu Gao, Yunpeng Zhang, Zhixia Li, Yongxin Liu, Tanvir Arafin (2025). International Conference on Transportation and Development 2025.
3. Safety and Longevity-Enhanced Energy Management of Fuel Cell Hybrid Electric Vehicle With Machine Learning Approach. Ruoyang Song, Xinghua Liu, Zhongbao Wei, Fengwen Pan, Yanbo Wang, Hongwen He (2024). IEEE Transactions on Transportation Electrification.
4. MLGO: a Machine Learning Guided Compiler Optimizations Framework. Mircea Trofin, Yundi Qian, E. Brevdo, Zinan Lin, K. Choromanski, D. Li (2021). arXiv.org.
5. Optimizing Machine Learning Operators and Models for Specific Hardware Using Apache-TVM. Kausthub Thekke Madathil, Abhinav Dugar, Nagamma Patil, Unnikrishnan Cheramangalath (2023). International Conference on Computing Communication and Networking Technologies.
6. BladeDISC: Optimizing Dynamic Shape Machine Learning Workloads via Compiler Approach. Zhen Zheng, Zaifeng Pan, Dalin Wang, Kai Zhu, Wenyi Zhao, Tianyou Guo, Xiafei Qiu, Minmin Sun, Junjie Bai, Feng Zhang, Xiaoyong Du, Jidong Zhai, Wei Lin (2023). Proc. ACM Manag. Data.