Efficient Algorithms for General Isotone Optimization
Algorithmic work on general order-constrained optimization, focused on efficient methods that scale beyond the narrowest special cases.
AAAI · 2022Xiwen Wang · Hong Kong
Xiwen Wang
I received my Ph.D. from HKUST after completing my undergraduate studies at Nanjing University. My work centers on large-scale optimization algorithms, machine learning, and embedded image processing.
Beyond research, this site keeps a visual record of my appetite for seeing the world: travel routes, photographs, notes, and moving skies from places I have been and places I still want to understand.
Motion
Yellowknife, over time
A few aurora timelapses from the northern end of the travel map: light, weather, and waiting compressed into short loops.
Selected Photos
Small fragments from a wider route
These are a few frames I would keep near the front of the house: Banff and Yellowknife for the northern light, Kobe and Japan for quiet city textures, Jiuzhaigou, Nanjing, and Xi'an for the long familiar route, then New York, Barcelona, and Luzern for the western edge of the map.
Research
Optimization and computation
My academic work centers on optimization algorithms, numerical computation, and structured models that stay useful at practical scale.
Efficient and Scalable Parametric High-Order Portfolios Design via the Skew-t Distribution
A portfolio-optimization formulation that brings higher-order moments into a scalable parametric model instead of treating them as a purely theoretical extension.
IEEE Transactions on Signal Processing · 2023Learning Large-Scale MTP2 Gaussian Graphical Models via Bridge-Block Decomposition
A decomposition-based route for learning large-scale structured graphical models while keeping the computation practical at nontrivial problem sizes.
NeurIPS · 2023Methods I keep returning to
- Active Set Methods
- ADMM
- Block Coordinate Descent
- Fixed Point Acceleration
- Majorization-Minimization
- Projected Gradient Methods
- Quasi-Newton Methods
- Successive Convex Approximation
For papers, collaborations, or related conversations, the direct route is contact.