Congratulations to Dr. Mengtian Guo!

VACLab-er Mengtian Guo successfully defended her dissertation in Fall 2025 and officially graduated with her PhD in Information Science in December. Her dissertation, titled “Facilitating Human-Machine Teaming in Machine Learning Problem Formulation,” addresses a fundamental yet often overlooked challenge in applied machine learning: how to better support practitioners as they decide what to predict and how to define it given the data they have available.

More specifically, Dr. Guo’s research focuses on the problem formulation stage of the machine learning pipeline, in which data scientists must bridge the gap between a high-level prediction goal and the concrete, measurable proxy variables that can actually be extracted from real-world data. This process typically involves significant trial-and-error and is poorly supported by existing tools—yet its outcomes can have a major impact on the quality and relevance of the resulting model. Dr. Guo investigated how human-machine teaming can help address this gap through two complementary studies. The first compared a machine-led strategy, in which the system recommends formulations based on predictive performance, against a human-led strategy, in which practitioners select proxy variables based on domain relevance. The second study extended this work by evaluating the effects of recommending Pareto-optimal problem formulations, treating the task as a multi-objective optimization problem that balances performance and relevance simultaneously. Across both studies, Dr. Guo found that introducing automation allowed users to examine a broader range of candidate formulations—but also revealed a consistent bias toward quantitative model quality metrics, which can lead to choices that are misaligned with the actual goals of the application. Based on these findings, Dr. Guo argues that future problem formulation tools must explicitly support the evaluation of user objectives and facilitate multi-criteria tradeoffs, not just raw predictive performance.

Dr. Guo has shared her work through a number of publications at leading venues, including the ACM Transactions on Interactive Intelligent Systems, Computer Graphics Forum (EuroVis), and ACM FAccT. She has now joined Google as a Software Engineer, bringing her expertise in human-machine collaboration and interactive machine learning systems to new challenges in industry.

Congratulations to the newest VACLab alumna, Dr. Mengtian Guo!