Contact

Address: Room 514, Hans Freudenthal building, Utrecht University, Budapestlaan 6, 3584 CD Utrecht, The Netherlands.

Email: z.huang1@uu.nl

CV

Education

• PhD candidate, Mathematical Institute of Utrecht University (Feb 2022 - until now)
  • Topic: Numerical Methods for Stochastic Control problems and FBSDEs
  • Promotor: Prof. Kees Oosterlee
• Master of Science in Financial Engineering, The Chinese University of Hong Kong (Shenzhen)
• Bachelor of Economics, Guangzhou University

Work experience

• Teaching Assistant, The Chinese University of Hong Kong (Shenzhen), Dec 2018 - Jan 2022

A complete CV can be obtained upon request.

Miscellaneous

• Google Scholar
• Math Stackexchange
• Wolfamalpha online tools
• Geogebra

Notes

ADMM

Background: ADMM (Alternating Direction Method of Multipliers) was proposed 40 years ago and recently attracted lots of attention. The convergence of 2-block ADMM for convex problems was known; however, a recent paper in 2014 showed that multi-block ADMM can diverge even for solving a 3*3 linear system. Interestingly, if we randomly permuted the update order in each cycle (e.g. (132), (231),… compared to traditional cyclic order (123), (123),…), then the algorithm converges. The question is: why?

Our contribution:

1. Result. We show that for solving linear systems RP-ADMM (randomly permuted ADMM) converges in expectation for any number of blocks.
2. High-level idea. One simple explanation for this phenomenon is “symmetrization’’: the update matrix of cyclic ADMM is a non-symmetric matrix with complex eigenvalues, and random permutation partially symmetrize the update matrix to make the eigenvalues have a nicer distribution. In fact, the key result is that the eigenvalues of the update matrix of RP-CD (randomly permuted coordinate descent) lies in (-1/3, 1), a smaller region than that of cyclic CD which is(-1,1).
3. Implications.
   1. Problem level: RP-ADMM can potentially be a good solver for large-scale linearly constrained problems (LP, SDP, etc.)
   2. Algorithm level: It was widely believed that RP rule is better than cyclic rule; however, little theory is established. This work provides one of the first theoretical results that RP rule is better than the cyclic rule.
      • On the Expected Convergence of Randomly Permuted ADMM Ruoyu Sun, Zhi-Quan Luo, Yinyu Ye.

Matrix Completion

–Updated 07/2016. Highlight the local geometry nature of our proof. New slides with a cleaner summary of the proof sketch.

Background: Motivated by applications such as recommender systems (e.g. Netflix prize), the problem of recovering a low-rank matrix from a few observations has been popular recently. It is a prototype example of how to utilize the low-rank structure to deal with big data. There are two popular approaches to impose the low-rank structure: nuclear norm based approach and matrix factorization (MF) based approach. The latter approach is especially amenable for big data problems, and has served as the basic component of most competing algorithms for Netflix prize. However, due to the non-convexity, it seems to be difficult to obtain a theoretical guarantee.

Publications

Published in , 1900

Matrix Completion

–Updated 07/2016. Highlight the local geometry nature of our proof. New slides with a cleaner summary of the proof sketch.

Background: Motivated by applications such as recommender systems (e.g. Netflix prize), the problem of recovering a low-rank matrix from a few observations has been popular recently. It is a prototype example of how to utilize the low-rank structure to deal with big data. There are two popular approaches to impose the low-rank structure: nuclear norm based approach and matrix factorization (MF) based approach. The latter approach is especially amenable for big data problems, and has served as the basic component of most competing algorithms for Netflix prize. However, due to the non-convexity, it seems to be difficult to obtain a theoretical guarantee.

ddd

Research

“If people do not believe that mathematics is simple, it is only because they do not realize how complicated life is.”

— John von Neumann

Research Interests

My research interests lie at the intersection of numerical methods and applied probability, with applications in finance, such as stochastic control, (F)BSDEs, stochastic modeling in financial markets and risk management, and scientific machine learning.

Publications

• Zhipeng Huang, and Cornelis W. Oosterlee. Convergence of the Markovian iteration for coupled FBSDEs via a differentiation approach. arXiv preprint arXiv:2504.02814 (2025)

• Negyesi, Balint, Zhipeng Huang, and Cornelis W. Oosterlee. Generalized convergence of the deep BSDE method: a step towards fully-coupled FBSDEs and applications in stochastic control. arXiv preprint arXiv:2403.18552v2 (2025) arXiv preprint arXiv:2403.18552v2 (2025)

• Zhipeng Huang, Balint Negyesi, and Cornelis W. Oosterlee. Convergence of the deep BSDE method for stochastic control problems formulated through the stochastic maximum principle. Mathematics and Computers in Simulation 227 (2025): 553-568. https://doi.org/10.1016/j.matcom.2024.08.002

• Michael CH Choi, and Zhipeng Huang. Generalized Markov chain tree theorem and Kemeny’s constant for a class of non-Markovian matrices. Statistics & Probability Letters 193 (2023): 109739. https://doi.org/10.1016/j.spl.2022.109739

Supervisions at Utrecht University

• BSc Thesis: Henrik van de Langemheen, Feb - Jul, 2025
  Topic: The COS method for computing the first hitting time of polynomial processes

• MSc Seminar Project: Jochem Lange and Martijn Brouwer, Feb - Jun, 2024
  Topic: Uncovering the underlying dynamics from data using machine learning techniques

Teaching

, , 1900

“The purpose of life is to discover your gift; the work of life is to develop it; and the meaning of life is to give your gift away.”

— David Viscott

Teaching Assistant at Utrecht University

• 2024 Fall: Numerical Methods for Stochastic Differential Equations, instructor: Dr. Chiheb Ben Hammouda
• 2024 Spring: Seminar on Neural Networks and Finance, organized by Prof. Kees Oosterlee

Teaching Assistant at CUHK-SZ

Prior to my PhD studies, I worked at CUHK-Shenzhen for several years as a full-time teaching assistant, primarily supporting courses in probability, statistics, and financial engineering. In 2021, I was awarded the Presidential Exemplary Teaching Assistant Award (less than 1%).

• MFE5110 Stochastic Models (Fall 2021), instructor: Prof. Nan Chen (CUHK)
• MFE5150 Financial Data Analysis (Spring 2021), instructor: Prof. Lingfei Li (CUHK)
• STA4020 Statistical Modeling in Financial Market (Fall 2020, 2021), instructor: Dr. John Wright (CUHK)
• MAT3007 Optimization 1 (Summer 2020), instructor: Prof. Andre Milzarek
• DDA6010 Optimization Theory (Fall 2019), instructor: Prof. Stark Draper (University of Toronto)
• STA4001 Stochastic Process (Fall 2019), instructor: Prof. Jim Dai (CUHK-SZ & Cornell)
• RMS4060 Risk Management with Derivatives (Summer 2019), instructor: Prof. Sang Hu
• STA2001 Probability and Statistics 1 (Spring 2019, 2020, 2021), instructor: Prof. Tianshi Chen
• DDA6001 Stochastic Process (Spring 2019, 2020, 2021), instructor: Prof. Masakiyo Miyazawa (CUHK-SZ & Tokyo University of Science)