Efficient LLM
Advancing compact and practical LLM architectures while preserving their expressive intelligence.
I study how to build efficient yet powerful LLMs for real-world practicality.
I am a Ph.D. candidate in Artificial Intelligence at Ajou University, advised by Prof. Jongbin Ryu. My work sits at the intersection of natural language processing and systems: architecture compression, on-device inference and acceleration, and hyperdimensional computing as a hardware-friendly substrate for learning.
Two papers accepted to EMNLP 2026: Layer-wise Curriculum Learning for Efficient LLM Compression and Train Overcomplete, Deploy Compact.
Submitted a patent on layer-wise curriculum learning for efficient LLM compression.
Appointed to an instructor for Efficient LLM Research program.
Started the Ph.D. program in Artificial Intelligence at Ajou University, advised by Prof. Jongbin Ryu.
Received a B.S. in Electronic Engineering from Ajou University.
How can we build efficient yet powerful AI for real-world practicality?
Advancing compact and practical LLM architectures while preserving their expressive intelligence.
Enabling powerful and energy-efficient AI processing directly on resource-constrained edge devices.
Restructuring AI computation through high-dimensional/parallel, brain-inspired processing.
† denotes the corresponding author. Names in bold are mine.
@inproceedings{lee2026layerwise,
title = {Layer-wise Curriculum Learning for Efficient {LLM} Compression},
author = {Lee, Donggeon and Na, Dooyeon and Oh, Seungmin and Ryu, Jongbin},
booktitle = {Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
year = {2026}
}
@inproceedings{oh2026train,
title = {Train Overcomplete, Deploy Compact: Scaling Recovery Capacity for Structured {LLM} Pruning},
author = {Oh, Seungmin and Lee, Donggeon and Ryu, Jongbin},
booktitle = {Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
year = {2026}
}
Ajou University · Advisor: Prof. Jongbin Ryu
Ajou University
I am happy to talk about efficient LLMs, on-device inference, hyperdimensional computing, or potential collaborations. The fastest way to reach me is email: donggeon_lee@ajou.ac.kr.