Department of Computer Science and Engineering · Inha University
Large-scale Machine Learning Systems Lab.
The Large-scale Machine Learning Systems Lab. (LMLS Lab) is a research group in the department of Computer Science and Engineering at Inha University, South Korea.
Our Mission
We study numerical optimization algorithms and systems for training machine learning models at scale across many heterogeneous machines and devices. Our work spans three connected threads: 1) distributed/federated learning, where we build optimizers that effectively utilize distributed data; 2) system-efficient deep learning, where we develop compute-, communication-, and memory-efficient learning methods that bridge the gap between theory and practice; 3) applied machine learning, where we study practical use-cases of various machine learning algorithms for scientific applications such as Physics.
We are always interested in hearing from self-motivated students and collaborators — see the People and Contact pages for more.
Latest News
All news →- Sunwoo Lee was recognized as a Top Reviewer and awarded a complimentary full registration for NeurIPS 2026.
- Daniel and Junhyuk's paper, Ghosted Layers: Unconstrained Activation Alignment for Recovering Layer-Pruned LLMs has been accepted to NeurIPS 2026 main conference.
- LMLS-Lab members have won the Grand Prize at Inha AI challenge 2026! Huge congratulations!
- Yunjae Jeong has joined LMLS-Lab as an undergrad intern. Welcome!
- Daniel's paper, Locality-aware Redundancy Pruning for LLM Depth Compression has been accepted to EMNLP main conference.
Selected Publications
All publications →- Ghosted Layers: Unconstrained Activation Alignment for Recovering Layer-Pruned LLMs
- Locality-aware Redundancy Pruning for LLM Depth Compression
- ZOO-Prune: Training-Free Token Pruning via Zeroth-Order Gradient Estimation in Vision-Language Models
- GEM: A Scale-Aware and Distribution-Sensitive Sparse Fine-Tuning Framework for Effective Downstream Adaptation
- Layer-wise Update Aggregation with Recycling for Communication-Efficient Federated Learning
- Layer-Wise Adaptive Gradient Norm Penalizing Method for Efficient and Accurate Deep Learning
- Layer-wise Adaptive Model Aggregation for Scalable Federated Learning
On-going Projects
- Leading Generative AI Human Resources Development
- Mid-career Researcher Program (Creative Research)
- BK21 FOUR: Program for Fostering Future-generation Researchers