Hom-PGD+: Fast Reparameterized Optimization over Non-convex Ball-Homeomorphic Set
ICML 2026.
LIANG Enming 梁恩明
Research Assistant Professor
City University of Hong Kong
My research lies at the intersection of machine learning and optimization, with a focus on methods that respect hard constraints and scale to real decision-making systems. My work is motivated by applications in power grids, mobility systems, and climate resilience.
Please feel free to contact me if you want to discuss possible collaborations or research directions.
Selected Research
Learning and Optimization with Constraints
Hom-PGD+: Fast Reparameterized Optimization over Non-convex Ball-Homeomorphic Set
Gauge Flow Matching: Efficient Constrained Generative Modeling over General
Convex Set and Beyond.
Fast Projection-Free Approach (without Optimization Oracle) for Optimization
over Compact Convex Set.
Efficient Bisection Projection to Ensure Neural-Network Feasibility for
Optimization over General Set.
ICML 2025.
Homeomorphic Projection to Ensure NN Solution Feasibility for Constrained
Optimization.
Learning and Optimization Theory
Ride-Sourcing Optimization
Learning to Dispatch and Reposition on a Mobility-on-Demand Platform.
DFF: Decision-Focused Fine-tuning for Smarter Predict-then-Optimize with
Limited Data.
A Smart Predict-then-Optimize Framework for Vehicle Rebalancing Problem.
Transportation Research Part B. 2026
Tested in DiDi (largest ride-sourcing platform in China).
Tested in DiDi (largest ride-sourcing platform in China).
An Integrated Reinforcement Learning and Centralized Programming Approach for
Online Taxi Dispatching.
IEEE TNNLS 2021.
Power Grid Operation
Partially Permutation-Invariant Neural Network for Solving Two-Stage
Stochastic AC-OPF Problem.
IEEE TPWRS 2025.
Solving Chance-Constrained AC-OPF Problems by Neural Network with
Bisection-based Projection.
ACM E-energy 2025, EnergySP Workshop.
AI for Optimal Power Flow Tutorial.
Climate Change AI Summer School 2024 & 2026.
Traffic Signal Control
Research Overview
My recent work develops homeomorphism-based methods for constrained learning and optimization, including Homeomorphic Projection, Homeomorphic Optimization, and Gauge Flow Matching. These methods aim to make neural models efficient and reliable under constraints.
Background
I received my Ph.D. from the Department of Data Science at City University of Hong Kong, supervised by Prof. Minghua Chen. My Ph.D. thesis received the 2026 ACM SIGEnergy Doctoral Dissertation Award Honorable Mention. I received my B.Eng. from SYSU, advised by Prof. Renxin Zhong.
I have worked on the DeepOPF project with Prof. Steven Low, focusing on machine learning methods for power-grid operation. I also visited the University of Cambridge to work with Prof. Srinivasan Keshav on power-grid resilience under extreme weather, and contributed to the AI for Optimal Power Flow tutorial at the Climate Change AI Summer School with Prof. Priya L. Donti. I also collaborate with DiDi on machine learning methods for urban mobility-on-demand systems. Previously, I was a research intern at MSRA (Beijing, 2022) and Noah's Ark Lab (Shenzhen, 2021), working on reinforcement learning and machine learning for logistics and wireless optimization.