Jiwon Song
Ph.D. Student @ SNU VLSI Lab
Electrical and Computer Engineering, Seoul National University
Building efficient AI systems across models, hardware, and deployment.
Research Focus
My research focuses on efficient AI systems, including model compression, long-context inference, and hardware-aware acceleration for practical deployment.
News
Recent updates.
π CompactAttention accepted to NeurIPS 2026 as a Spotlight!
π Joined Amazon as an Applied Scientist Intern!
Publications
Research on efficient inference, reasoning acceleration, and practical AI system optimization.
Rotation-Aligned Key Channel Pruning for Efficient Vision-Language Model Inference
arXiv preprint arXiv:2605.19218
LiteStage: Latency-aware Layer Skipping for Multi-stage Reasoning
arXiv preprint arXiv:2510.14211
Education
Ph.D. Student (Integrated M.S./Ph.D. Program)
Electrical and Computer Engineering, Seoul National University
Advisor: Jae-Joon Kim
Bachelor of Science
Electrical and Computer Engineering, Seoul National University
Summa Cum Laude 3.9/4.3
Experiences
Applied Scientist Intern
Sep 2026 -- Alexa+
Research Intern
Summer 2022- DNN quantization
Technical Writer
2024- Blog posts comparing vLLM and TensorRT-LLM
- vLLM vs TensorRT-LLM #6: Weight-Only Quantization
- vLLM vs TensorRT-LLM #8: KV Cache Quantization
Undergraduate Researcher
2021-2022, 2023-2024Seoul National University
- LLM quantization and pruning
- Post-training quantization of vision transformers
Teaching Assistant
Spring 2024Dept. of Electrical and Computer Engineering, Seoul National University
- 430.207B κΈ°μ΄μ μνλ‘ λ° μ€ν (Introduction to Electronic Circuits)
Military Service
2020-2021Republic of Korea Army
- Sergeant