Distributionally Robust Optimization
Generative ambiguity modeling for optimization systems that need to perform under distribution shift and uncertainty.
Ph.D. Student · Computer Science · University of Houston
Jiaqi Wen
My research focuses on robust and trustworthy machine learning systems, with current work on generative distributionally robust learning, decision-focused learning, goal recognition, and causality.
Advisor & Lab
I am a Ph.D. student advised by Dr. Jianyi Yang in the Computational IntelliGence (CIG) Lab @ UH, where the team studies responsible and efficient AI/ML, decision making, and AI computing.
Research
Building reliable learning systems for uncertain environments.
Decision-Focused Learning
Learning methods that connect prediction quality with downstream decisions, including 3D and diffusion-augmented settings.
Goal Recognition & Causality
Variational and causal approaches for long-term goal recognition and interpretable AI reasoning.
3D Perception Systems
Point-cloud segmentation, measurement, recognition, and automated evaluation systems used in commercial deployments.
News
Recent updates on papers, collaborations, and research milestones.
I started a Robust RL collaboration with Professor Shi's team at Johns Hopkins University, and I am leading the project.
I am honored to serve as a reviewer for NeurIPS 2026.
We released WaterAdmin: Orchestrating Community Water Distribution Optimization via AI Agents on arXiv.
Our paper Distributionally Robust Optimization via Generative Ambiguity Modeling was accepted by ICLR 2026.
Our paper 3D-Learning: Diffusion-Augmented Distributionally Robust Decision-Focused Learning was accepted by IEEE INFOCOM 2026.
Our paper Distributionally Robust Optimization via Diffusion Ambiguity Modeling was accepted by NeurIPS OPT 2025.
Goal Recognition via Variational Causality appeared at AAMAS 2025.
Spectral-Pointer Network was published at CVIDL & ICCEA 2022.
Publications
Recent papers and accepted work.
ICLR
Distributionally Robust Optimization via Generative Ambiguity Modeling
J. Wen and J. Yang. ICLR 2026.
This work studies DRO through generative ambiguity sets that model adversarial distributions beyond the nominal support while preserving consistency with the nominal data. It introduces GAS-DRO and demonstrates stronger out-of-distribution generalization with a diffusion-model implementation.
NeurIPS OPT
Distributionally Robust Optimization via Diffusion Ambiguity Modeling
J. Wen and J. Yang. NeurIPS OPT 2025.
This paper designs diffusion-based ambiguity sets for DRO, enabling adversarial distributions outside the nominal support while remaining data-consistent. The proposed D-DRO algorithm improves out-of-distribution generalization in machine learning prediction tasks.
INFOCOM
3D-Learning: Diffusion-Augmented Distributionally Robust Decision-Focused Learning
J. Wen, L. Fan, and J. Yang. INFOCOM 2026.
3D-Learning addresses out-of-distribution failures in predict-then-optimize systems by training predictors for worst-case downstream decision performance. It uses diffusion models to search realistic worst-case distributions and improves results on LLM resource provisioning.
arXiv
WaterAdmin: Orchestrating Community Water Distribution Optimization via AI Agents
J. Wen, P. Tang, S. Ren, and J. Yang.
WaterAdmin combines LLM-based community context abstraction with optimization-based water-system control. The bi-level framework adapts pump and valve scheduling under dynamic community contexts and improves pressure reliability and energy consumption in EPANET simulations.
AAMAS
Goal Recognition via Variational Causality
J. Wen and L. Amado. AAMAS 2025.
This paper introduces a model-free goal recognition method that integrates causality through variational inference. The approach combines causal discovery, counterfactual inference, and trajectory-likelihood decision making, remaining robust under noisy observations.
CVIDL
Spectral-Pointer Network: Pre-sort Leads the Pointer Network to Elude the TSP Vortex
J. Wen. CVIDL & ICCEA 2022.
This work studies Pointer Network behavior on TSP-style optimization and proposes a spectral pre-sorting strategy to improve search stability and help the model avoid difficult tour-construction patterns.
Education
Interdisciplinary training across AI, computer science, and health science.
Ph.D. in Computer Science
University of Houston · USA
MSc in Artificial Intelligence
University of Aberdeen · Distinction · UK
MPhil in Computer Science
North China Electric Power University · China
BSc in Health Inspection and Quarantine
Guangdong Pharmaceutical University · China
Experience
Research, engineering, and production AI systems.
Research Assistant · University of Houston
uh.eduDoctoral student advised by Dr. Jianyi Yang, working on generative distributionally robust learning across theoretical foundations and practical applications.
Research Assistant · University of Aberdeen
abdn.ac.ukFocused on researching applications of causal inference and goal recognition. Contributed to long-term goal recognition research with Prof. Felipe and Dr. Amado.
Algorithm Engineer · MetaDigital
metadigital.net.cnDeveloped 3D laser point-cloud segmentation, building measurement, floor-plane recognition, and distributed cluster systems used at CRLAND, in collaboration with teams from Stony Brook University and the University of Tokyo.
Test Development Engineer · Goodix Technology
goodix.comLed TOF 3D sensor testing work, built point-cloud evaluation systems, contributed to depth-map completion models, and developed automation for face recognition and sensor testing. Responsible for testing mass production tools for fingerprint recognition MCUs.
Skills
Technical Stack
Python
PyTorch
TensorFlow
Open3D
NumPy
Pandas
scikit-learn
Matplotlib
OpenCV
MATLAB
SPSS
SAS
Awards
Selected Honors
- 2025 NSM Graduate Student Conference Travel Award
- 2023 Aberdeen Global Scholarship
- 2018 Honor of Outstanding Worker
Contact
