Sukmin Yun

I am currently a postdoctoral fellow at MBZUAI, working with Professor Eric Xing.

Previously, I was a Ph.D. student in the School of Electrical Engineering at Korea Advanced Institute of Science and Technology (KAIST), advised by Prof. Jinwoo Shin. 

My research revolves around realistic machine learning problems with limited supervision. This includes many practical and challenging scenarios, e.g., learning with noisy labels, (open-set) semi-supervised learning, and self-supervised learning. Specifically, my previous research has been focused on developing a new way to use more supervisory signals from model or data. Recently, I am also interested in multi-modal learning (e.g., Vision-Language). 

Contact: sukmin.yun [at] kaist [dot] ac [dot] kr

[Google Scholar] [Twitter]

Publications

(C: conference / W: workshop / J: journal / P: preprint) (* = equal contribution)

[C6] IFSeg: Image-free Semantic Segmentation via Vision-Language Model

[C5] Time Is MattEr: Temporal Self-supervision for Video Transformers (code)

[C4] TSPipe: Learn from Teacher Faster with Pipelines (code)

[C3] Patch-level Representation Learning for Self-supervised Vision Transformers (code) 

[C2] Regularizing Class-wise Predictions via Self-knowledge Distillation (code) 

[C1] Robust Inference via Generative Classifiers for Handling Noisy Labels (code) 

[W1] OpenCoS: Contrastive Semi-supervised Learning for Handling Open-set Unlabeled Data (code)

eDUCATION

Korea Advanced Institute of Science and Technology (KAIST)  

Korea Advanced Institute of Science and Technology (KAIST) 

wORK EXPERIENCE

External Research Collaborator

Teaching Assistant

Research Assistant

HONORS & AWARDS

Winner, Qualcomm Innovation Fellowship Korea 2022

Participation Grant, International Conference on Machine Learning (ICML) 2022

Excellent Paper Award, KAIST-Samsung Electronics Industry-University Cooperation, 2019

Travel Award, International Conference on Machine Learning (ICML) 2019

Academic Services

Conference Reviewer

Journal Reviewer