Minseo Kim

CS PhD Student at UC Berkeley

Hi, I’m a CS PhD student at UC Berkeley, advised by Prof. Kurt Keutzer. I’m broadly interested in efficient inference and AI systems. My recent interests include hybrid attention, post-training, and agentic systems. During my undergrad years, I worked with Prof. Kurt Keutzer and Dr. Amir Gholami at BAIR, and Prof. Jae W. Lee at SNU’s ARC Lab.

prof_pic.png

news

Nov 19, 2026 We released 2DSD, our work on two-dimensional SD for looped language models (LMs). It works with any block drafter and supports both model-wise and block-wise looped LMs.
Sep 24, 2026 Three papers (EfficientRollout, AsyncOPD, MAGE) accepted to NeurIPS 2026. Thanks to my collaborators at FuriosaAI, Berkeley, and SNU!
Aug 19, 2026 Starting my PhD journey at UC Berkeley 🎓

selected publications

  1. NeurIPS
    efficientrollout.png
    EfficientRollout: System-Aware Self-Speculative Decoding for RL Rollouts
    Minseo Kim*, Minjae Lee*, Seunghyuk Oh, and 7 more authors
    NeurIPS 2026, ICML 2026 AdaptFM Workshop, 2026
  2. ISCA MLArchSys
    llminaflash.png
    LLM Inference in a Flash!
    Sebastian Zhao*, Minseo Kim*, Coleman Hooper*, and 5 more authors
    ISCA 2026 MLArchSys Workshop (Oral), 2026
  3. MLSys
    cdlm.png
    CDLM: Consistency Diffusion Language Models for Faster Sampling
    Minseo Kim, Chenfeng Xu, Coleman Hooper, and 5 more authors
    In Proceedings of the 9th MLSys Conference, 2026
  4. beyond.png
    Beyond Next-Token Prediction: A Performance Characterization of Diffusion versus Autoregressive Language Models
    Minseo Kim, Coleman Hooper, Aditya Tomar, and 5 more authors
    arXiv Preprint , 2025