About Me

Hi! I’m an MSc student at KAIST AI, advised by Seong Joon Oh. I’m broadly interested in how LLMs can stay reliable and personalized as the knowledge they depend on keeps changing. Specifically, my research interests center on:

  • Retrieval-augmented generation (RAG) and retrieval methods
  • Knowledge editing through RAG, keeping models’ factual knowledge up to date
  • Building memory systems and personalization for long-term, persistent agents

Before my graduate studies, I spent six years co-founding and building a deep learning startup, working end-to-end across business, research, and engineering. That experience shaped how I think about research: not just as a technical problem, but as something that has to hold up in the messy, real-world systems where people actually rely on it.

In short, I’d like to build LLM-based systems that can remember, retrieve, and update what they know in a way that is both trustworthy and personalized to the people who use them.

Download my CV (PDF)

🔥 Recent News

  • [Oct 2026] Released Cross-Pollination Retrieval: Finding Transferable Ideas Across Fields in Scientific Literature as a preprint on arXiv.
  • [Sep 2026] Form Over Content in Gradient-Based Data Attribution Methods was accepted at the NeurIPS 2026 Workshop on Attributing Model Behavior at Scale (ATTRIB)!
  • [Sep 2026] Released TULIP: Targeted LLM Unlearning at Layers Identified Per-Input as a preprint on arXiv.
  • [Sep 2026] MEME: Multi-Entity & Evolving Memory Evaluation was accepted at NeurIPS 2026 (Evaluations & Datasets Track)!
  • [Sep 2026] Released Form Over Content in Gradient-Based Data Attribution Methods as a preprint on arXiv.
  • [Jul 2026] Our workshop proposal PALM: Personalized, Aligned, Long-term Memory for AI Systems was accepted at NeurIPS 2026. I’m serving as an organizer!
  • [May 2026] Paper accepted at the ICML 2026 Workshop on Scalable Learning and Optimization for Efficient Multimodal AI Agents (SCALE)!
  • [May 2026] Paper accepted at the ICML 2026 Workshop on the Impact of Memorization on Trustworthy Foundation Models!
  • [May 2026] Released MEME: Multi-Entity & Evolving Memory Evaluation as a preprint on arXiv.
  • [Feb 2026] Joined the Scalable Trustworthy AI (STAI) group at KAIST AI.
  • [May 2019] Paper accepted at ICASSP 2019.

📝 Selected Publications

* Equal contribution  ·  † Mentored intern

  1. MEME: Multi-Entity & Evolving Memory Evaluation Paper Code Project Page
    Seokwon Jung, Alexander Rubinstein*, Arnas Uselis*, Sangdoo Yun, and Seong Joon Oh
    Thirty-Ninth Conference on Neural Information Processing Systems (NeurIPS 2026), Evaluations & Datasets Track
  2. Cross-Pollination Retrieval: Finding Transferable Ideas Across Fields in Scientific Literature Paper
    Seokwon Jung, Yerang Kim†, Yejin Kim, Alexander Rubinstein, Arnas Uselis*, Sangwoo Park*, Seong Joon Oh, and Sangdoo Yun
    arXiv preprint arXiv:2610.25678
  3. TULIP: Targeted LLM Unlearning at Layers Identified Per-Input Paper
    Yejin Kim, William F. Shen, Seokwon Jung, Daeun Park, and Seong Joon Oh
    arXiv preprint arXiv:2609.34591
  4. Form Over Content in Gradient-Based Data Attribution Methods Paper
    Sunwoo Kim, Seokwon Jung, SoHyung Kim, Seong Joon Oh, and Alice Oh
    NeurIPS 2026 Workshop on Attributing Model Behavior at Scale (ATTRIB)
  5. Break the Output Geometry for Large Language Model Unlearning Paper
    Yejin Kim, William F. Shen, Seokwon Jung, and Seong Joon Oh
    ICML 2026 Workshop on the Impact of Memorization on Trustworthy Foundation Models
  6. Polyphonic Sound Event Detection Using Convolutional Bidirectional LSTM and Synthetic Data-based Transfer Learning Paper
    Seokwon Jung, Jungbae Park, and Sangwan Lee
    ICASSP 2019 - 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)