ICRA 2023 · Paper
Road Anomaly Segmentation Based on Pixel-wise Logit Variance with Iterative Background Highlighting
Using pixel-level uncertainty to identify unexpected hazards in road scenes.
Senior Staff Engineer
Dongkun Lee.
AI systems. LLM inference. Model compression.
I develop algorithms that make AI models more efficient, connecting research with practical deployment.
KAIST
01 / Selected work
Research & projects
ECAI 2023 · Paper
A Simple Debiasing Framework for Out-of-Distribution Detection in Human Action Recognition
Reducing bias to improve the detection of unfamiliar actions in video.
AI application · Project
AILA - AI Language Assistant
A conversational learning platform connecting AI dialogue with interactive 3D avatars.
02 / Background
From research to deployment
Current role
Senior Staff Engineer
Huawei · 2025–present
I work on LLM inference optimization and model efficiency, balancing model quality, memory, latency, and throughput.
Full experience & educationResearch foundation
Ph.D., KAIST
My research spans out-of-distribution detection, anomaly segmentation, uncertainty estimation, and computer vision, with publications at ICRA and ECAI.
Technical focus
- LLM inference
- Quantization
- Pruning & sparsity
- Distillation
- Python / PyTorch
03 / Contact
Let’s exchange ideas.
Research exchange and open-source collaboration in efficient AI.