Bio

I’m a Master of Science student in Statistical Science at Duke University, with a strong background in Computer Science and Statistics from Wake Forest University. My research focuses on High Performance Computing (HPC), Efficient and Scalable Machine Learning, Numerical Methods, and other key subjects. As a Research Assistant at Duke’s CEI Lab, I work on interdisciplinary projects combining statistics, computer science, and applied mathematics.


Selected Publications

SADA Project

SADA Project

SADA: Stability-guided Adaptive Diffusion Acceleration
Ting Jiang, Hancheng Ye*, Yixiao Wang*, Zishan Shao, Jingwei Sun, Jingyang Zhang, Jianyi Zhang, Zekai Chen, Yiran Chen, Hai Li
ICML 2025 Poster

Abstract: Diffusion models have shown remarkable success in generating high-quality samples, but their inference process is often slow due to the need for many denoising steps. We propose SADA (Stability-guided Adaptive Diffusion Acceleration), a novel approach that adaptively reduces the number of denoising steps while maintaining sample quality. Our method uses a stability-guided criterion to determine when to skip steps, achieving significant speedup without compromising generation quality.

Paper | Code | Slides

ECCD Project

ECCD Project 1 ECCD Project 2

Enhanced Cyclic Coordinate Descent for Elastic Net Penalized Linear Models
Zishan Shao, Yixiao Wang, Ting Jiang, Aditya Devarakonda
NeurIPS 2025 (Submitted)

Abstract: Coordinate descent methods are widely used for solving large-scale optimization problems, particularly in machine learning applications. We propose Enhanced Cyclic Coordinate Descent (ECCD), an improved variant that addresses the limitations of traditional cyclic coordinate descent for Elastic Net penalized linear models. Our method introduces adaptive step sizes and efficient communication-avoiding strategies, achieving faster convergence and better scalability for high-dimensional problems.

Paper | Code | Slides

FlashSVD Project

ECCD Project

FlashSVD: Memory Efficient Approach for SVD-Based Low Rank Model Inference
Zishan Shao, Yixiao Wang, Qinsi Wang, Ting Jiang, Zhixu Du, Hancheng Ye, Danyang Zhuo, Yiran Chen, Hai Li
AAAI 2026 (Submitted)

Abstract: Singular Value Decomposition (SVD) is a fundamental technique in machine learning and data analysis, but its memory requirements can be prohibitive for large-scale applications. We present FlashSVD, a memory-efficient approach for SVD-based low-rank model inference that significantly reduces memory footprint while maintaining accuracy. Our method employs innovative caching strategies and adaptive memory management, enabling efficient processing of large matrices on memory-constrained systems.

Paper | Code | Slides


Education

  • Duke University, Durham, NC

    • Master of Science in Statistical Science, Expected 2025
    • Advisor: Dr. Yiran Chen
    • Research Lab: CEI Lab (Computational Engineering and Intelligence)
    • Relevant Coursework: Real Analysis, Advanced Numerical Linear Algebra, Numerical Methods, Bayesian Statistics, Advanced Statistical Inference, Predictive Modeling & Statistical Learning, Electrical Computing in Neural Networks, Advanced Stochastic Process, Advanced Hierarchical Modeling.

  • Wake Forest University, Winston-Salem, NC

    • Bachelor of Science in Computer Science, May 2024
    • Second Major (B.S.) in Statistics
    • Dean’s List: All semesters, Summa Cum Laude
    • Dissertation: “Communication Avoiding Coordinate Descent Methods On Kernel And Regularized Linear Models” PDF
    • Advisor: Dr. Aditya Devarakonda
    • Relevant Coursework: Python, C, R, Java, Statistical & Machine Learning, Statistical Inferences, Linear Models, Multivariate Analysis, Probability, Network Theory & Analysis, Advanced Data Structures and Algorithms, Assembly Languages, Computer System, Multivariate Calculus, Linear Algebra, Discrete Mathematics, Time Series Forecasting, Special Topics – Computer Vision, High-Performance Computing, Parallel Computing

  • Shanghai Weiyu High School, Shanghai, China

Academic Profiles