Hongseok Namkoong

Assistant Professor, Columbia University

I am an assistant professor in the Decision, Risk, and Operations division and Computer Science (affiliate), and a member of the Data Science Institute. I work on building trustworthy AI systems that are capable of continually improving itself through interactions with the world. I take a data-centric view of AI systems, and am a strong believer in algorithmic ideas simultaneously grounded in empirical foundations and principled thinking. As an interdisciplinary researcher, I connect and extend tools from machine learning, operations research, and statistics. Read my research foundations overview for a summary of my work. For what is most top of mind these days, see my latest agenda on lifetime-learning agents. I expect prospective students and postdocs who want to work with me to have read these before reaching out.

Before joining Columbia, I received my Ph.D. from Stanford University and spent a year at Meta’s Adaptive Experimentation team as a research scientist. Outside of academia, I served as a LinkedIn Scholar at LinkedIn’s Core AI team. Here’s a more formal bio in the third person. I go by Hong; alternatively, here’s a link the correct pronunciation of my first name.

news

Jun 23, 2026 MBABench: Evaluating LLMs on Spreadsheet Tasks Across Critical Dimensions
Oct 14, 2025 I’m looking for motivated undergraduate and master’s students to work on ML research. Fill this form out if you’re interested.
Dec 05, 2024 AI systems are omni-present, yet extrapolate unreliably. Improving AI safety and capabilities hinges on comprehension of uncertainty and actively making decisions to resolve it. Instead of cumbersome probabilistic models, my team leverages a predictive view of uncertainty to build a scalable framework based on autoregressive models. Watch this recent Simons talk to learn more.

selected publications

  1. MBABench: Evaluating LLM Agents on End-to-End Spreadsheet Tasks in Finance
    Trustworthy AI
    Thomson Yen, Julian Poeltl, Harshith Srinivas Gear, Yilin Meng, and 8 more authors
    arXiv:2605.22664 [cs.AI], 2026
    arXiv Bib Code Data Website
  2. LatentGym: A Testbed for Cross-Task Experiential Learning with Controllable Latent Structure
    Trustworthy AI
    Daksh Mittal, Tommaso Castellani, Thomson Yen, Naimeng Ye, and 6 more authors
    arXiv:2606.15306 [cs.LG], 2026
    arXiv Bib Blog Code Data
  3. Active Exploration via Autoregressive Generation of Missing Data
    AI-driven Decisions
    Tiffany Cai, Hongseok Namkoong, Daniel Russo, and Kelly Zhang
    arXiv:2405.19466 [cs.LG], 2025
    Selected for presentation at the Econometric Society Interdisciplinary Frontiers: Economics and AI+ML conference
    arXiv Bib Slides