Geunwoo Kim

I am a Ph.D. student in Computer Science at UC Irvine, under the supervision of Michael Franz and Pierre Baldi.

I interned in Coupang's search team, where I enhanced the retrieval relevance via large-scale graph neural networks. Additionally, I worked closely with Dokyung Song on graph representation learning for binary code. Prior to this, I obtained my B.S. in Computer Science from Pohang University of Science and Technology, where I researched blockchain security with Jong Kim. In 2018, I also served as the president of the Postech Laboratory for Unix Security, a university hacking team. During my undergraduate break, I spent six months as a blockchain developer at Kodebox in Gangnam, Seoul, followed by another six months as a research assistant at the National University of Singapore. There, I worked with Min Suk Kang on blockchain network layer security.

Email  /  LinkedIn  /  Github

profile photo
Research

My current research interests revolve around the development and application of efficient machine learning systems.

Deep Learning Assisted Imaging Methods to Facilitate Access to Ophthalmic Telepathology
Geunwoo Kim*, Andrew W. Browne*, Anderson N. Vu, Josiah K. To, Don S. Minckler, Maria Del Valle Estopinal, Narsing A. Rao, Christine A. Curcio, Pierre F. Baldi
Ophthalmology Science 2023

We propose employing super-resolution imaging techniques, specifically utilizing the Denoising Diffusion Probabilistic Model (DDPM), to create high-fidelity pathology slide images. This approach offers a cost-effective alternative to advanced digital slide scanners for remote telepathology services using commonly available smartphones.

Language Models can Solve Computer Tasks
Geunwoo Kim, Pierre Baldi, Stephen McAleer
NeurIPS 2023
AI & HCI workshop @ ICML 2023
project page / arXiv

A pre-trained large language model agent can execute computer tasks guided by natural language using a simple prompting scheme, recursively criticizing and improving its output.

Improving Cross-Platform Binary Analysis using Representation Learning via Graph Alignment
Geunwoo Kim, Sanghyun Hong, Michael Franz, Dokyung Song
International Symposium on Software Testing and Analysis (ISSTA) 2022
project page / video / paper

We propose XBA that uses a semisupervised approach to generate binary code embeddings which are aligned across platforms by collecting peripheral information of binary code using graph neural network.

The Ticket Price Matters in Sharding Blockchain
Geunwoo Kim, Michael Franz, Jong Kim
International Workshop on Cryptocurrencies and Blockchain Technology (ESORICS workshop) 2022
code / paper

We explore the impact of non-democratic environments on the security and scalability of blockchain sharding, proposes metrics to analyze the trade-off between security and scalability.

This webpage template was recycled from here