Hyeonwoo Noh

Email: hyeonwoonoh [at] gmail [dot] com
Google scholar page

I’m doing research on artificial intelligence.

Previously I was a research lead at OpenAI. I founded and led the Computer-Using Agent team, pioneering computer-use capabilities in OpenAI’s models powering the Operator and the ChatGPT Agent. Before that, I led visual understanding research for GPT-4, enabling OpenAI’s models to process images and bringing image understanding to ChatGPT. I finished my Ph.D. at the Pohang University of Science and Technology under supervision of Suha Kwak and Bohyung Han.

I am interested in building machines that can solve a variety of emerging problems by constantly acquiring skills and learning to understand the problems. My research focuses on learning deep neural networks with a few general interfaces to solve a large number of problems.

Education

Ph.D. Computer Science and Engineering, Pohang University of Science and Technology (March, 2015 - July, 2019)
B.S. Computer Science and Engineering, Pohang University of Science and Technology (March, 2007 - July, 2013)

Industrial Experience

Research Scientist. OpenAI, United States (Sept, 2019 - Jan, 2026)
Research Intern. DeepMind, United Kingdom (July, 2018 - Dec, 2018)
Research Intern. Google, United States (June, 2016 - Dec, 2016)
Software Engineer. Samsung Electronics, South Korea (July, 2013 - Jan, 2015)

Research

ChatGPT Agent

OpenAI In 2025

Operator / Computer-Using Agent

OpenAI In 2025

GPT-4 [blog post] / ChatGPT can now see

OpenAI In 2023

Asymmetric self-play for automatic goal discovery in robotic manipulation [project page]

OpenAI In 2021

Behavior Priors for Efficient Reinforcement Learning

Dhruva Tirumala, Alexandre Galashov, Hyeonwoo Noh, Leonard Hasenclever, Razvan Pascanu, Jonathan Schwarz, Guillaume Desjardins, Wojciech Marian Czarnecki, Arun Ahuja, Yee Whye Teh, Nicolas Heess
In JMLR 2023

Real-Time Object Tracking via Meta-Learning: Efficient Model Adaptation and One-Shot Channel Pruning

Ilchae Jung, Kihyun You, Hyeonwoo Noh, Minsu Cho, Bohyung Han
In AAAI 2020

Exploiting Hierarchy for Learning and Transfer in KL-regularized RL

*Dhruva Tirumala, *Hyeonwoo Noh, Alexandre Galashov, Leonard Hasenclever, Arun Ahuja, Greg Wayne, Razvan Pascanu, Yee Whye Teh, Nicolas Heess
In Arxiv 2019 (*Equal contribution)

Transfer Learning via Unsupervised Task Discovery for Visual Question Answering

Hyeonwoo Noh, Taehoon Kim, Jonghwan Mun, Bohyung Han,
In CVPR 2019 [code]

Neural Program Synthesis from Diverse Demonstration Videos

*Shao-Hua Sun, *Hyeonwoo Noh, Sriram Somasundaram, and Joseph Lim,
In ICML 2018 (*Equal contribution) [project page]

Regularizing Deep Neural Networks by Noise: Its Interpretation and Optimization

Hyeonwoo Noh, Tackgeun You, Jonghwan Mun, and Bohyung Han,
In NIPS 2017 [poster][video]

Large-Scale Image Retrieval with Attentive Deep Local Features

Hyeonwoo Noh, Andre Araujo, Jack Sim, Tobias Weyand, Bohyung Han,
In ICCV 2017 [code]

Training Recurrent Answering Units with Joint Loss Minimization for VQA

Hyeonwoo Noh, Bohyung Han,
In Arxiv [project page]

Image Question Answering using Convolutional Neural Network with Dynamic Parameter Prediction (oral presentation)

Hyeonwoo Noh, Paul Hongsuck Seo, Bohyung Han,
In CVPR 2016 [project page]

Decoupled Deep Neural Network for Semi-supervised Semantic Segmentation (spotlight presentation)

*Seunghoon Hong, *Hyeonwoo Noh, Bohyung Han,
In NIPS 2015 (*Equal contribution) [project page]

Learning Deconvolution Network for Semantic Segmentation

Hyeonwoo Noh, Seunghoon Hong, Bohyung Han,
In ICCV 2015 [project page]

Seam Carving with Forward Gradient Difference Maps

Hyeonwoo Noh, Bohyung Han,
In ACM Multimedia(MM) 2012