(Adam) Cheol Woo Kim

I am a postdoctoral fellow in Computer Science at Harvard University, advised by Professor Milind Tambe. My research advances the scientific foundations of end-to-end automated decision-making using machine learning. My work spans multiple layers of the decision-making process, embedding machine learning throughout the pipeline from aligning with human intent to solving optimization problems and incorporating human feedback.

• AI Alignment and LLMs for Decision-Making
  I develop methods to align learning-based systems with human objectives, particularly in settings where preferences are ambiguous, multi-faceted, or uncertain. My research spans both aligning LLMs with human values and leveraging LLMs for decision-making:
  1. Multi-objective alignment: Designing frameworks that balance multiple human objectives and enable personalized policy selection through human-in-the-loop methods.
  2. Robust fine-tuning: Developing robust algorithms for LLM fine-tuning that explicitly account for noise and uncertainty in human preference data.
  3. LLMs for decision-making: Exploring how LLMs can assist in complex decision-making tasks, such as translating natural language descriptions into reward functions for reinforcement learning.
• Learning to Optimize
  I develop ML-accelerated algorithms that enable real-time solutions to complex optimization and control problems. My work spans mixed-integer optimization, two-stage robust optimization, and continuous-time optimal control, achieving orders-of-magnitude speed-ups while establishing theoretical optimality guarantees.

Before joining Harvard, I completed my PhD at the MIT Operations Research Center in 2024 under the guidance of Professor Dimitris Bertsimas. I also spent the summer of 2023 as a research intern in the Machine Learning and Optimization Group at Microsoft Research. I completed my undergraduate studies at Seoul National University.