Minseo Kim

Undergraduate Student, CSE, Seoul National University

prof_pic.png

Hi, I’m an undergraduate student at Seoul National University majoring in Computer Science and Engineering. My research focuses on improving the efficiency of large-scale AI models grounded in a deep understanding of systems. I aim to achieve this by developing system-aware algorithmic methods and cross-stack designs that enable efficient training and serving in real-world deployments. Currently, I am working on inference efficiency for large language models (LLMs) and diffusion language models (DLMs).

During my undergraduate years, I have been fortunate to be part of two great research groups. I am currently a visiting researcher in the Pallas Lab at Berkeley AI Research (BAIR), advised by Prof. Kurt Keutzer and Dr. Amir Gholami. Previously, I worked in the Architecture and Code Optimization Lab (ARC Lab) at Seoul National University, advised by Prof. Jae W. Lee.

I am seeking a PhD position starting in Fall 2026.

news

Nov 24, 2025 Our paper on accelerating DLM inference via fine-tuning is now on arXiv. Huge thanks to my collaborators - Chenfeng, Coleman, and Harman! [Link]
Nov 14, 2025 Team Architects won the Grand Prize (NIPA President’s Award) at the 2025 AI Chip Contest!
Oct 07, 2025 Our paper on DLM analysis is now on arXiv. [Link] This is my first paper at Berkeley! :sparkles:
Aug 21, 2025 Our VLM team at AttentionX had a paper accepted to EMNLP 2025! I’ll be presenting it in Suzhou, China (Nov 5-7). [Link]

selected publications

  1. cdlm.png
    CDLM: Consistency Diffusion Language Models for Faster Sampling
    Minseo Kim, Chenfeng Xu, Coleman Hooper, and 5 more authors
    Under review for MLSys , 2025
    arXiv Code
  2. 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
    arXiv
  3. EMNLP
    kreta.png
    KRETA: A Benchmark for Korean Reading and Reasoning in Text-Rich VQA Attuned to Diverse Visual Contexts
    Taebaek Hwang*, Minseo Kim*, Gisang Lee, and 2 more authors
    In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 2025
    DOI HTML Code
  4. CIKM
    instanns.png
    InstANNS: Scalable Approximate Nearest Neighbor Search via Cost-Efficient In-Storage Processing
    Bonggeun Sim, Yushin Kim, Minseo Kim, and 2 more authors
    In Proceedings of the 34th ACM International Conference on Information and Knowledge Management (CIKM ’25), 2025
    DOI HTML