Dongwon Jo

I am a Ph.D. student in Electrical & Computer Engineering at Seoul National University, advised by Prof. Jae-Joon Kim. I received my B.S. in Electrical & Electronic Engineering from Yonsei University. My research interests are summarized below.

Research Interests

My primary research aims to make model compression broadly applicable in real-world scenarios. A central theme of my work focuses on reducing the memory and compute costs of Deep Neural Networks (DNNs) that achieve superior accuracy through increased complexity, via algorithm–hardware co-design. Specifically, my research agenda includes:

• Development of DNN accelerators
• Design of hardware-friendly DNNs (e.g., quantized or sparsified models)
• Model compression algorithms for efficient inference

Currently, my research primarily targets generative models, including large language models (LLMs) and diffusion models, with a focus on practical efficiency and scalability. As these models are increasingly deployed in long-context settings, the associated memory and compute costs grow rapidly, creating fundamental barriers to real-world adoption. My work seeks to address these challenges through principled algorithm design, with the goal of making powerful generative models accessible under realistic resource constraints. Ongoing research topics include:

• Quantization and pruning algorithms for LLMs and diffusion models
• KV cache compression and sparse attention for long-context LLM inference
• Kernel-level optimization for high-throughput generative models

*Keywords: Generative Models, Efficient Inference, Model Compression, Algorithm-Hardware Co-design

Selected Publications [Full List]

Token Sparse Attention: Efficient Long-Context Inference with Interleaved Token Selection

Dongwon Jo, Beomseok Kang, Jiwon Song, Jae-Joon Kim

ICML 2026

Paper Code Project

FastKV: Decoupling of Context Reduction and KV Cache Compression for Prefill-Decoding Acceleration

Dongwon Jo*, Jiwon Song*, Yulhwa Kim, Jae-Joon Kim

ACL Findings 2026

Paper Code

Mixture of Scales: Memory-Efficient Token-Adaptive Binarization for Large Language Models

Dongwon Jo, Taesu Kim, Yulhwa Kim, Jae-Joon Kim

NeurIPS 2024

Paper Code

Education

Work Experiences

Academic Services

Teaching