Ryan (Feng) Lin

My long-term research vision is to develop a generic methodological framework to build personalized digital twins for many real-world applications where individual heterogeneity is crucial, such as healthcare, transportation, and business. To realize this vision, there are three essential pillars: 1) Human-centric AI. Humans, as core participants who contribute data and define the success of personalized digital twins, introduce significant variation and uncertainty with their heterogeneous nature. It is essential for AI to understand human behaviors and foster effective interaction. I have led a research team developing a novel framework called CrowdLLM that builds on Large Language Models and generative AI, and can incorporate human behavior models. I have also developed models to characterize how human interact with AI systems and how rewards can be personalized to incentivize behavioral changes for improving individual well-being and system-wide performance. 2) Domain knowledge. While domain knowledge offers a stable foundation grounded in human experience and scientific discovery, the key challenge lies in extracting, representing and operationalizing such knowledge to effectively inform personalized digital twins. Addressing this challenge is essential for enhancing their realism, interpretability, and controllability. 3) AI systems. As digital twins are envisioned to be high-fidelity virtual representation of the physical twin, there are many methodological gaps in machine learning and AI model that range from modeling, to uncertainty quantification, and calibration, etc. My research have been focusing on these three pillars and a synergistic cycle among these three is the main drive of my research.

My research is interdisciplinary and lies at the intersection of artificial intelligence (AI), statistics, and quality science, with applications spanning healthcare, medicine, transportation, marketing, etc. I have been collaborating with researchers from both academia and industry across the United States, Europe, and Asia. I was also fortunate to work at Meta as a full-time AI researcher for one year, supported by a generous Meta AI Fellowship, and gained precious experience of cutting-edge AI research and their applications in the real-world. My research works that focused on AI model personalization, human diversity and behavior, their relevance with LLMs and generative AI, have been selected as finalists in the Best Paper Competition of INFORMS Quality, Statistics, and Reliability (QSR) Section in 2023 and 2025. A summary of my research is demonstrated in the overview figure.

My research has been supported by a variety of funding sources including government agencies (e.g., NSF, DARPA), non-profit organizations (e.g., Breakthrough T1D, formerly JDRF), and industry sponsors (e.g., Amazon, Meta). My research also aligns well with the priorities of NSF programs such as SCH, FDT-BioTech, MATH-DT, Future CoRe (formerly SCC), etc., ARPA-H initiatives including GIVE and THRIVE, as well as several NIH programs. In addition, agencies such as DARPA, USDOT, or the U.S. Army regularly issue solicit research relevant to my expertise. I also intend to explore new partnerships with industry (e.g., Microsoft, NVDIA) or non-profit organizations. Looking ahead, I will build a strong and sustainable research program supported by these funding sources.