Sungbin Shin

prof_pic.jpeg

Hi, I am a Ph.D student in computer science and engineering at POSTECH, advised by Prof. Namhoon Lee.

I am broadly interested in large-scale optimization for foundation models, with a focus on distributed and collaborative learning. I explore the algorithmic foundations that enable efficient, scalable training of billion-scale models across diverse participants, developing principled methods that push the boundaries of scalable machine learning.

Contact me via sungbin.shin@postech.ac.kr if you have any questions.

news

Dec 2026 ✈️ I will be attending NeurIPS 2026 in Sydney! If you’d like to chat about research, feel free to reach out.
Nov 2026 🎖️ I have been awarded a MARS Visiting Fellowship at Lancaster University! In February–March 2027, I will spend a month at the School of Mathematical Sciences working with Dr. Mher Safaryan on communication- and memory-efficient optimization algorithms. Check out the official announcement!
Nov 2026 ✈️ I’ll be in Casablanca to attend ICOMP. Feel free to reach out if you’re interested in large-scale optimization for foundation models!
Jul 2026 🎖️ I have been selected for the PhD Excellence Scholarship, a nationally competitive scholarship for outstanding Ph.D. students in science and engineering jointly funded by the Ministry of Science and ICT and the Korea Student Aid Foundation (KOSAF)
Jul 2026 ✈️ I will be presenting our paper on basis rotation for asynchronous pipeline parallelism at ICML 2026 in Seoul, including at the ICML Workshop: Protocol Learning organized by Pluralis Research. Come find me if you’re around, or feel free to reach out!

Earlier news can be found here.

selected publications

2026

  1. Mitigating Staleness in Asynchronous Pipeline Parallelism via Basis Rotation
    Hyunji Jung*, Sungbin Shin*, Namhoon Lee
    ICML, 2026

2025

  1. Critical Influence of Overparameterization on Sharpness-aware Minimization
    Sungbin Shin*, Dongyeop Lee*, Maksym Andriushchenko, Namhoon Lee
    UAI, 2025; ICML Workshop on High-dimensional Learning Dynamics, 2023
    Best paper award at JKAIA 2023

2024

  1. Rethinking Pruning Large Language Models: Benefits and Pitfalls of Reconstruction Error Minimization
    Sungbin Shin, Wonpyo Park, Jaeho Lee, Namhoon Lee
    EMNLP, 2024

2023

  1. A Closer Look at the Intervention Procedure of Concept Bottleneck Models
    Sungbin Shin, Yohan Jo, Sungsoo Ahn, Namhoon Lee
    ICML, 2023; NeurIPS Workshop on TSRML, 2022