Sihyun Yu

sihyun.yu at kaist dot ac dot kr
yusihyunc at gmail dot com

I am a Ph.D. student at Algorithmic Intelligence Laboratory @ KAIST, advised by Jinwoo Shin. I am working closely with Kihyuk Sohn at Google Research. Prior to this, I received B.S. in Computer Science and Mathematics (double major) at KAIST under the supervision of Hongseok Yang and Juho Kim, who kindly advised me to be a conscientious reseacher.

News

Nov 2022: I will attend NeurIPS 2022 offline!
Sep 2022: “Scalable Neural Video Representations with Learnable Positional Features” is accepted to NeurIPS 2022.

Research

My research focuses on designing memory-/computation-efficient generative models. In particular, I am interested in solving this challenge by constructing generative models with various structural assumptions and regularizations in latent space. Currently, I am actively working on video synthesis, but not limited to; other complex data (e.g., 3D scenes and graphs) and applications are also in my interest. Besides, I also have an interest in the robustness of neural networks to the distributional shift.

2022

Scalable Neural Video Representations with Learnable Positional Features
Subin Kim*, Sihyun Yu*, Jaeho Lee, Jinwoo Shin
NeurIPS 2022
paper | project page | code

Generating Videos with Dynamics-aware Implicit Generative Adversarial Networks
Sihyun Yu*, Jihoon Tack*, Sangwoo Mo*, Hyunsu Kim, Junho Kim, Jung-Woo Ha, Jinwoo Shin
ICLR 2022
paper | project page | slides | code

Consistency Regularization for Adversarial Robustness
Jihoon Tack, Sihyun Yu, Jongheon Jeong, Minseon Kim, Sung Ju Hwang, Jinwoo Shin
AAAI 2022
paper | slides | code

Learning Robust Representations via Nuisance-extended Information Bottleneck
Jongheon Jeong, Sihyun Yu, Hankook Lee, Jinwoo Shin
ECCV 2022 Workshop on Out of Distribution Generalization in Computer Vision (OOD-CV)
paper

2021

RoMA: Robust Model Adaptation for Offline Model-based Optimization
Sihyun Yu, Sungsoo Ahn, Le Song, Jinwoo Shin
NeurIPS 2021
paper | slides | code

Abstract Reasoning via Logic-guided Generation
Sihyun Yu, Sangwoo Mo, Sungsoo Ahn, Jinwoo Shin
ICML 2021 Workshop on Self-Supervised Learning for Reasoning and Perception (oral)
paper | slides | poster

Collaborations

Google Research
with Dr. Kihyuk Sohn as Google University Relations (Jul 2022 - present)

NAVER AI Lab
with Dr. Jung-Woo Ha as an external collaborator (Jun 2021 - Sep 2021)

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)
with Prof. Le Song and Prof. Sungsoo Ahn as an external collaborator (Mar 2021 - May 2021)

Korea Advanced Institute of Science and Technology (KAIST)
with Prof. Hongseok Yang (April 2018 - June 2019)

Honors and Awards

Top Reviewer, NeurIPS, 2022

Best Paper Awards, Korean Artificial Intelligence Association, 2021

Recipient, Qualcomm-KAIST Innovation Awards, 2019

Recipient, KAIST Presidental Fellowship, 2018-2020

Recipient, National Presidental Scholarship for Science, 2017-2020

Recipient, Hansung Scholarship for Gifted Students, 2015-2016

Invited Talks

RoMA: Robust Model Adaptation for Offline Model-based Optimization
Jun 2022; Samsung Electronics (Suwon, South Korea)

Scaling Video Generation via Implicit Neural Representations
Jun 2022; Pohang University of Science and Technology (Pohang, South Korea)

Abstract Reasoning via Logic-guided Generation
Jun 2021; ICML Workshop on Self-Supervised Learning for Reasoning and Perception 2021 (remote)

Academic Services

Conference reviewer: ICML'{22,23}, NeurIPS'22, ICLR'23, CVPR'23

Workshop reviewer: AI4CC@CVPR'22, Neural-Fields@ICLR'23

Teaching

Teaching Assistant, Deep Learning (AI502), Fall 2020

Teaching Assistant, Samsung Electronics Device Solutions, AI-Expert Program, Summer 2020

Peer Tutor, System Programming (CS230), Spring 2019

Teaching Assistant, Linear Algebra (MAS212), Spring 2019

Teaching Assistant, TA, Calculus 1 (MAS101), Spring 2018