(3jp) Small-Data Learning for Sustainable Chemical Engineering

My research aims to accelerate the development of innovative and sustainable chemical engineering technologies during the early design stages, with the overarching goal of supporting the transition toward a more sustainable chemical industry. Machine learning holds great promise in this context, as it offers the ability to uncover complex patterns and relationships within industrial data. To fully realize this potential, however, we must overcome a critical challenge: the scarcity of high-quality data for novel chemicals and processes. My research focuses on addressing this gap to unleash the full power of data-driven design in sustainable chemical engineering.

During my Ph.D., I systematically developed data-efficient machine learning methods to mitigate data scarcity. (1) Transfer learning-aided feature selection. This approach leverages data from "similar" sources to enrich the target domain. At the same time, feature selection helps reduce task complexity by filtering out irrelevant variables, which is particularly valuable when data availability is limited. By effectively characterizing task similarity, we identified optimal hyperparameter tuning strategies that maximize model accuracy. These strategies account for both the number of data points and dissimilarity between tasks, offering a principled way for users to adapt and apply the method to a variety of specific cases. (2) Integration of prior knowledge. We developed an efficient data-mining algorithm capable of automatically extracting logical rules from limited industrial data. This prior-knowledge-informed model mitigates overfitting and enhances predictive accuracy by embedding domain-relevant insights into the data-driven learning process. These methods have demonstrated remarkable performance in improving the silicon content prediction of hot metal from blast furnaces and medical monitoring applications, particularly with the Parkinson’s telemonitoring dataset.

To address this challenge in greater depth, my current postdoctoral research aims to bridge the data gaps in the life cycle assessment (LCA) of organic chemical production, a particularly data-scarce area, by designing fully automated predictive tools that integrate chemical engineering knowledge with machine learning to predict large-scale life cycle inventories (LCIs) from molecular structure solely. By transparently modeling synthesis pathways and LCIs, our tools can be flexibly adapted to various LCA databases, life cycle impact assessment (LCIA) methods, and system models. Therefore, our tools allow for filling data gaps in LCA databases for early-stage process design and accelerate the transition toward a sustainable chemical industry.

My future work will continue to focus on developing trustworthy, data-efficient algorithms capable of overcoming data limitations and serving as a strong bridge between advanced machine learning / mathematical techniques and chemical engineering applications. These applications include, but are not limited to, the development of new life-cycle-based inventory and impact assessment methods for improved environmental sustainability assessment of organic chemical and metal-organic framework (MOF) production with a particular focus on CO2 capture and storage (CCS) technologies.

Teaching Interests

Aligned with my research interests, my teaching interests include introductory courses on Life Cycle Assessment (LCA) and CO2 capture and storage (CCS) technologies. Additionally, given their methodological relevance to LCA and CCS, I am also interested in teaching foundational and applied courses in Machine Learning, Statistics, Linear Algebra, and Mathematical Analysis.

My teaching philosophy is based on the belief that learning is an interactive process, one that depends on both clear, accessible explanations from the instructor and active curiosity and critical thinking from the student. I believe students learn best when vivid, real-world examples are used to spark their interest and help them connect classroom concepts to everyday experiences. As a teaching assistant for the graduate-level course “CO2 Capture and Storage and the Industry of Carbon-Based Resources” at ETH Zurich, I had the opportunity to design and present exercise sessions on the topics related to my research. In these sessions, I emphasized connecting theoretical content to practical, real-life examples, this approach was well-received by the students. I also served as a teaching assistant for the undergraduate-level course “Linear Algebra” at Zhejiang University. There, I received positive feedback for helping students grasp abstract mathematical concepts by illustrating their relevance through real-life applications and connections across different areas of mathematics.