Jinuk Kim

Hello! I am a 3rd year PhD student at Seoul National University, Computer Science department, Machine Learning Lab, advised by Hyun Oh Song. My research interests lie in constructing efficient machine learning system by solving tractable discrete and continuous optimization problem. I received my Bachelor's degree in Statistics from Seoul National University in 2023.

CV   /  Scholar   /  Github   /  Twitter   /  Blog   /  LinkedIn

• May 2025 One paper got accepted in ICML 2025 (GuidedQuant).

• Aug 2024 I will be joining Google as a Student Researcher.

• May 2024 One paper got accepted in ICML 2024 (LayerMerge).

• Apr 2024 I have been awarded the Presidential Science Scholarship for Graduate Students (120/2980=4.0%).

• Aug 2023 I will be joining Samsung Advanced Institute of Technology as a Research Intern.

• May 2023 One paper got accepted in ICML 2023 (Efficient-CNN-Depth-Compression).

• May 2022 One paper got accepted in ICML 2022 (Efficient-Dataset-Condensation).

• GuidedQuant proposed an improved objective function and quantization method for large language models.

• LayerMerge and Efficient-CNN-Depth-Compression introduced novel depth compression methods to accelerate inference of CNN and diffusion model.

• Efficient-Dataset-Condensation improved the efficiency of neural network training via dataset condensation.

We propose GuidedQuant, a novel quantization approach that integrates gradient information from the end loss into the layer-wise quantization objective. Additionally, we introduce LNQ, a non-uniform scalar quantization algorithm which is guaranteed to monotonically decrease the quantization objective value.

We propose LayerMerge, a novel depth compression method that selects which activation layers and convolution layers to remove, to achieve a desired inference speed-up while minimizing performance loss.

We propose a subset selection problem that replaces inefficient activation layers with identity functions and optimally merges consecutive convolution operations into shallow equivalent convolution operations for efficient inference latency.

We propose a novel condensation framework that generates multiple synthetic data with a limited storage budget via efficient parameterization considering data regularity and develop an effective optimization technique.

Android/iOS service which collects notices from website of SNU departments and gather them (Android / Aug 2021 / 100+ MAU / 80+ WAU / 1000+ Downloads).

• Conference Reviewer: TMLR 2024-, ICML 2025-, NeurIPS 2025-

Last updated: May 15, 2025. Template based on Jon Barron's website.