Meet Haoruo Zhao, a PhD student and Operations Research and Machine Learning researcher with AI4OPT.

Research Focus and Accomplishments

I’m currently collaborating with Keysight on AI testing and verification in wireless communications. My work focuses on developing practical testing and verification methods to ensure AI-driven methods, such as deep learning for CSI prediction, perform accurately and efficiently in real-world conditions.

Previously, I worked on the RAMC project, where we aimed to replicate and enhance the MISO pipeline using digital twins. Our goal was to improve existing methods to better handle uncertainty and risk. My focus was on developing a stochastic version of the security-constrained economic dispatch problem. I also contributed to building a simulator covering the entire industry pipeline—from time series prediction to day-ahead and intra-day unit commitment, all the way to economic dispatch.

More recently, my research has shifted toward neural network verification. We built a benchmark dataset for verifying optimization proxies for ACOPF problems and developed efficient methods to ensure the reliability of these neural networks. Our work on the benchmark was recognized with the VNNCOMP 2023 Best Benchmark Award.

Career Development and Goals

I aspire to pursue industry research, developing innovative solutions to real-world challenges. I enjoy tackling complex problems and bridging the gap between theory and application, ensuring that the methods I work on are both effective and impactful. Whether in academia or industry, the most rewarding part of research for me is seeing ideas translate into meaningful solutions.

To achieve my career goals, I am actively developing my engineering skills while deepening my domain knowledge in operations research and machine learning. Through my research and industry collaborations, I’ve learned that strong engineering is essential for impactful ideas. Additionally, I leverage AI tools to stay updated with the latest research and improve efficiency. Most importantly, I am honing my ability to accurately understand real-world problems and build solutions that truly work—because the most valuable research is the kind that effectively solves the problem at hand.

Research Impact and Engagement

My research on stochastic look-ahead dispatch contributes to better uncertainty management in power systems, particularly in integrating renewable energy. The framework demonstrates the feasibility of using stochastic economic dispatch for real-time market clearing, improving efficiency and reliability. Additionally, my work in neural network verification introduces optimization proxies for power systems to the neural network verification community, encouraging the development of more advanced tools to verify these proxies and formally ensure the safety of critical applications.

Personal Interests

Outside of research, I enjoy hiking, spending time with my cat, and discovering great restaurants with friends. As the saying 民以食为天 (food is the most important to people) goes, I love exploring and trying different cuisines—a passion deeply rooted in my culture and a great way to unwind.

Fun Fact

For some real-life problems, I instinctively start thinking about the best solution right away. For example, when I see a restaurant delivering food to multiple destinations, I can’t help but think about the optimal route to minimize travel distance—essentially solving a real-world version of the Traveling Salesman Problem. This mindset naturally led me to operations research, where I apply mathematical optimization to complex decision-making challenges.

That said, solving everything optimally takes too much energy. Take grocery shopping, for example—I could carefully plan the most efficient route through the store to minimize walking distance. But instead, I just wander around, grab what I need… and a few things I probably shouldn’t have bought!

Publications

1. A Linear Outer Approximation of Line Losses for DC-based Optimal Power Flow Problems – Power Systems Computation Conference (PSCC 2022)
2. Bound Tightening Using Rolling-Horizon Decomposition for Neural Network Verification – International Conference on the Integration of Constraint Programming, Artificial Intelligence, and Operations Research (CPAIOR 2024)
3. Compact Optimality Verification for Optimization Proxies – International Conference on Machine Learning (ICML 2024)

Conference Presentations and Awards

I presented my work on stochastic look-ahead dispatch at the 2023 Grid Science Winter School and Conference, where I received the Best Poster Award.