Ph.D. Candidate in Computer Science and Engineering
University of Notre Dame · Expected May 2027
Faculty job market · Fall 2026
I am on the academic job market this fall, seeking faculty positions. CV (PDF) · Contact me
I am a Ph.D. candidate advised by Prof. Yiyu Shi at the University of Notre Dame. I pioneer hardware and compression methods that help AI work well for more people.
AI performance gaps have largely been treated as problems of data and algorithms. I established a new hardware research direction by showing that the computing hardware itself affects how well AI serves different groups (Nature Electronics, 2024).
I also give model compression a new purpose. My work shows that tools built to make AI smaller can help close these gaps, establishing new uses for familiar efficiency techniques. I am turning this research into AI assistants that help users put these ideas into practice.
🎉 It is my great honor to have received the 2026 Outstanding Graduate Student Teaching Award from Kaneb Center for Teaching Excellence and Graduate School, University of Notre Dame.
Discoveries that expand what AI hardware and model compression can do.
AI performance gaps have largely been treated as problems of data and algorithms. I revealed hardware design as another source of these gaps. This work established a new research direction for understanding and improving how AI serves different groups.
Compression has traditionally been used to make models smaller and cheaper to run. I pioneered a new use for quantization: reducing performance gaps between groups. FairQuantize turns an efficiency tool into a way to improve how AI serves people.
I introduced FairCompressAgent to turn these discoveries into an interactive design process. Users describe their goals; an AI assistant explores and tests ways to adapt their models. This brings a new research direction closer to practical use.
I am developing this new role for hardware and compression into a broader research program: understanding what causes performance gaps, finding new ways to close them, and making these methods easier to use.
Hardware imperfections are usually treated as problems to eliminate. Our FairXbar work showed that they can also be used to reduce performance gaps between groups. This turns a hardware limitation into a new opportunity for improving AI.
Building on the idea behind FairQuantize, I introduced FairLRF to give another compression tool a new purpose. It shows that breaking a model into simpler parts can reduce performance gaps between groups, extending this new direction beyond a single technique.
Choosing a compression method and adapting it to a new model often takes specialist expertise. FairCompressAgent introduces an AI assistant that turns user goals into options it can test and refine. My aim is to make this approach to reliable AI usable beyond its original research setting.
International Conference on Field Programmable Technology (ICFPT). Invited paper.
Design, Automation & Test in Europe Conference (DATE).
Acceptance rate: 24.8%
Nature Electronics, 7, 714–723.
Journal impact factor: 42.3
ACM Transactions on Design Automation of Electronic Systems.
Journal impact factor: 2.8
IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 1–9.
Acceptance rate: 23.5%
* Equal contribution.
Impact factors are journal-level metrics.
arXiv:2511.16549. Under review; revised September 2026.
2026 Outstanding Graduate Student Teaching Award
University of Notre Dame
Award announcement
Student evaluation scores: 5.0/5.0 for all completed teaching assistant semesters.
Fall 2026
Develop and grade laboratory assignments, support student group projects, and hold office hours.
Spring 2025 & 2026
Developed homework and final-exam materials, graded homework, quizzes, and exams, and held office hours across two course offerings.
Spring 2022
Graded homework and examinations and held office hours for a remote course.
Fall 2021
Graded homework, assisted with final-exam assessment, and held office hours.
Pingchuan Dong (2026): master’s student at Columbia University.
Joy Xie and Kevin Luo (2022–2024): mentored as high-school researchers; both co-authored our Nature Electronics paper. They were subsequently admitted to MIT and Stanford, respectively.
May–Dec. 2025
Independently evaluated image and language AI models on CPUs and dedicated AI hardware, connecting research ideas with the constraints of real deployment. Worked with the compiler team to support models on specialized hardware.
Oct. 2019–Jul. 2021
Contributed to CNN-Cap under Prof. Wenjian Yu, applying AI to predict electrical interactions between chip wires. The work showed how learning-based models could improve a task traditionally handled by conventional estimation methods, leading to publications in ICCAD and ACM TODAES.
Reviewer for Medical Image Computing and Computer Assisted Intervention (MICCAI, 2025) and Smart Health (2024).