Large-Scale Optimization Proxies
Learning fast solution mappings for large mixed-integer problems with strong temporal and system-wide coupling.
Georgia Tech ISyE · NSF AI4OPT
DongJun Lee
Ph.D. Student · Machine Learning & Optimization
Optimization-grounded machine learning for constrained sequential decision-making under uncertainty.
I develop fast, feasibility-preserving methods for large-scale decision systems whose actions are coupled over time and across system components.
Advised by Prof. Pascal Van Hentenryck
About
Constrained sequential decision-making under uncertainty.
I develop optimization-grounded machine learning methods for systems whose decisions are coupled over time and across system components, and must satisfy hard operational constraints. My work studies how learned models can propose or construct decisions while optimization structure and feasibility repair preserve temporal and system-wide consistency.
My current work develops full ML proxies for large-scale mixed-integer optimization and context-conditioned policies for multistage stochastic optimization. The proxy research targets instances with more than 100,000 discrete variables and millisecond-scale inference and repair. My earlier work in time-series forecasting and anomaly detection provides a foundation in long-range temporal modeling, multivariate dependency learning, and uncertainty-aware representation.
Research
Core directions
Learning methods for fast, feasible decisions in large-scale systems with temporal and system-wide coupling.
Decision Policies Under Uncertainty
Developing context-conditioned policies for multistage settings where information arrives over time and decisions must remain operationally valid.
Feasibility-Preserving Learning & Repair
Integrating optimization structure, structured repair, and deployment-aligned objectives into learned decision pipelines.
Sequential & Distributional Modeling
Long-range temporal representation, multivariate dependency modeling, scenario generation, and uncertainty calibration.
Selected Projects
Research in focus
Current and earlier research across optimization proxies, multistage decision policies, and sequential representation learning.
A Full ML Proxy for Large-Scale Mixed-Integer Optimization
Developing a full ML proxy for large-scale, temporally coupled mixed-integer optimization. The target instances contain more than 100,000 discrete variables, and the research aims to combine learned inference with structured feasibility repair at millisecond-scale latency while preserving system-wide operational constraints.
100K+discrete variables
ms-scaletarget latency
ML + repairfeasibility pipeline
Contextual Decision Policies for Multistage Stochastic Optimization
Developing context-conditioned decision policies for multistage stochastic optimization, with optimization-integrated training, feasibility-aware deployment, and stage-wise evaluation under the information available at each decision point.
View DecisionRules.jlPublications
Publications
Work across time-series forecasting, anomaly detection, and sequential representation learning.
* Equal contribution.
-
2025
-
2025
A Comprehensive Survey of Deep Learning for Time Series Forecasting: Architectural Diversity and Open Challenges
Artificial Intelligence Review
A structured review of deep-learning architectures for time-series forecasting and the open problems that shape reliable long-horizon prediction.
View publication -
2024
-
2016
A Study on the Posture Based Control of Robotic Fixation Device
International Journal of Applied Engineering Research, 11(9), 6655–6659
Early work on posture-based control for a robotic fixation device.
View journal issue
Applied Research / Experience
Applied research across real-world decision systems.
Industry collaborations and independent work across credit risk, mobility, batteries, markets, and industrial diagnostics.
- Mixed-integer optimization
- Stochastic optimization
- Sequential decision-making
- Feasibility repair
- Probabilistic forecasting
- Time-series learning
- Python
- Julia
Aug. 2025–Present
Graduate Researcher
Georgia Institute of Technology · NSF AI4OPT
Researching learning-augmented optimization and scalable decision policies for large, structured systems under the supervision of Prof. Pascal Van Hentenryck.
Jan.–Apr. 2026
Explainable Attention-Based Credit Risk Modeling
Equifax · Atlanta, USA
Built an end-to-end attention-based credit risk modeling framework and developed PCA/CCA-guided analyses connecting latent Transformer features to human-defined credit attributes.
Aug. 2025–Jan. 2026
Risk-Constrained Trading Agent
Independent Research
Developed a leakage-safe multi-asset data pipeline, a calibrated multi-task Transformer, and cost-aware backtests with volatility-scaled sizing, exposure limits, and drawdown controls.
Mar.–Jul. 2025
Optimization Proxies for Driving Performance
Hyundai · Seoul, South Korea
Automated optimal transmission-control calibration maps and developed a constraint-driven, physics-consistent model-to-vehicle matching pipeline.
Aug. 2024–Jun. 2025
EV Battery Defect Anomaly Detection
LG Energy Solution · Seoul, South Korea
Designed process-aware tokenization and contrastive learning for highly imbalanced, noisy charge-discharge data, then fused time-series embeddings with expert tabular features.
Mar.–Oct. 2024
AI-Based Motor Calibration
Hyundai Genesis · Seoul, South Korea
Developed deep-learning methods to predict motor calibration maps from sparse operating-point data for reducing sixth-harmonic torque ripple and noise.
Additional experience
Jul. 2023
Associate Consultant Intern
PricewaterhouseCoopers Consulting · Hwaseong, South Korea
Jun.–Jul. 2021
Data Scientist Intern
OnePredict · Seoul, South Korea
Education
Academic background
Georgia Institute of Technology · ISyE
Ph.D. Student, Industrial Engineering
NSF AI Institute for Advances in Optimization (AI4OPT)
Advisor: Prof. Pascal Van Hentenryck
Seoul National University
M.S., Artificial Intelligence
Data Science & Artificial Intelligence Laboratory (DSAIL)
Advisor: Prof. Sungroh Yoon
Seoul National University
B.S., Mechanical Engineering
Summa Cum Laude
Contact
Let’s discuss machine learning, optimization, and decision systems.
The most direct way to reach me is by email. You can also find my publications and current work through the profiles below.
djlee@gatech.edu