My original name is 岳乾, pronounced [ywè-tɕhɛň] in Mandarin. I also go by Lizzy. 😊

Contact

✉️ [email protected]

Research Interests

Machine Learning Systems

Computer Architecture

Software Hardware Co-design

Machine Learning

High-performance Computing

Languages

🇺🇸 English

🇨🇳 Mandarin

Awards

• Outstanding Graduate of Shanghai
• Outstanding Student Leader of Tongji University
• Second Prize in the National College Students' Mathematical Modeling Competition
• Third Prize in the Tongji University Mathematics Competition

<aside> <img src="/icons/cactus_orange.svg" alt="/icons/cactus_orange.svg" width="40px" /> In research, I am particularly drawn to the concept of software-hardware co-design. I am also passionate about addressing energy consumption in computing systems and aim to improve performance while reducing energy costs.

I consider myself a principled and organized collaborator who values systematic investigation, sound assumptions, rigorous proofs, and precise claims. I believe that articulating insights is essential for advancing scientific progress.

While I excel in structured environments, I recognize that my biggest challenge lies in managing chaotic materials (such as code, data, and ideas). In my experience, I thrive in collaborative settings with active thinkers, as this often leads to superior outcomes. 😉

</aside>

Education

Tongji University, School of Mathematics and Science

Sep. 2014 - Jul. 2018, GPA 93/100, with a B.S degree in science, Shanghai

Completed coursework in Optimization Theory, Mathematical Analysis, C/C++ Programming, Combinatorics and Graph Theory, among others.

National Sun Yat-sen University, Department of Applied Mathematics

Jan. 2016 - Jul. 2016, GPA 4.0/4.0, exchange student, Taiwan

Completed coursework in Complex Analysis, Statistics, Ordinary Differential Equation, among others.

Fudan University, School of Data Science

Sep. 2018 - Jul. 2020, GPA 3.52/4.0, with a M.S degree in the field of economics, Shanghai

Completed coursework in Computer Vision, Medical Image Analysis, Analytical and Managerial Economics, among others.

Work Experience

NIO, Senior Architecture Performance Engineer

Feb. 2023 - Apr. 2025

• Led architecture exploration for NIO's Neural Processing Unit (NPU), evaluating the feasibility of a 2D mesh interconnect to replace the hierarchical ring, achieving a 20% average improvement in data transmission efficiency.
• Developed and validated cycle-accurate C++ models of AI accelerators for autonomous driving, leading to a 30% reduction in simulation time and more accurate performance predictions.
• Designed over 400 highly optimized NPU kernels, boosting software efficiency and ensuring peak chip performance in real-world autonomous driving applications.
• Identified and resolved 10+ critical RTL bugs through high-level simulation, significantly improving the reliability of the chip design.
• Designed and implemented performance visualization tools for bus transmissions, enabling real-time analysis of the interconnect under diverse traffic scenarios.
• Enhanced core fairness and reduced head-of-line blocking by applying advanced techniques like virtual channels and credit-based flow control, leading to a 10% improvement in resource allocation.

Intel, AI Framework Engineer

Dec. 2021 - Feb. 2023

• Optimized GPU kernels using time and resource profiling, contributing to the successful submission of the BERT model for MLPerf, a widely recognized AI workload benchmark for chip manufacturers.
• Boosted performance of compute and memory-intensive kernels by analyzing Intel GPU hardware characteristics and utilizing advanced architectural features like DPAS (Dot Product Accumulate Systolic) and 2D loading.

Research Experience

Fudan University, School of Data Science, ZMIC Lab

Sep. 2018 - Sep. 2019

Yue Q, Luo X, Ye Q, et al. Cardiac segmentation from LGE MRI using deep neural network incorporating shape and spatial priors[C]//Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part II 22. Springer International Publishing, 2019: 559-567.

This research presents a novel approach to fully automated MRI segmentation, tackling the challenges posed by heterogeneous intensity distributions and indistinct boundaries in medical images.

By incorporating shape reconstruction and spatial information, our convolution neural network effectively delineates cardiac structures and employs these shapes as constraints during training. Consequently, the model achieves segmentation accuracy comparable to that of medical students, alleviating the workload for radiologists.

This work was published in MICCAI, a premier conference in medical image analysis. For visual results, please refer to the attached poster.

MICCAI - Poster.pdf

The Chinese University of Hong Kong, Shenzhen, as a visiting student

Jul. 2018 - Sep. 2018

• Participated in a smart campus project at The Chinese University of Hong Kong, leveraging big data technologies. Utilized large streams of student data, such as library access, course attendance, and cafeteria usage, to predict students' GPA for the current semester. The system provided tailored academic support and guidance to assist teaching staff and advisors. I was responsible for the theoretical proof of a key component of this research.
• Conducted theoretical research on the XGBoost algorithm. Proved that the quantile of ensemble tree classification results can be constrained by a confidence interval. Through theoretical analysis, I estimated the lower and upper bounds of this interval and conducted numerical experiments, which revealed that merging decision tree samples effectively reduces the length of the confidence interval, leading to more precise predictions and improving the overall accuracy and reliability of the model's performance.

More about Me