Lee Optimization Group
Oct 2026 🎓 Our preprint on decentralized LLM training, SeedFlood, has received a major update, with substantial improvements to its overall presentation, analysis, and experimental evaluation.
Oct 2026 🧑‍💻 Minkyoung has started a research internship at Nota AI.
Oct 2026 🎓 Our paper on hardware-native sparse-quantization for trillion-scale MoEs is out on arXiv. In this work, we introduce MoESQ, which jointly learns semi-structured sparsity and low-precision quantization for MoE experts and executes them on Sparse Tensor Cores with a custom grouped sparse GEMM kernel. The code and models are publicly available.
Sep 2026 🎓 Our paper on decentralized training has been accepted to NeurIPS 2026 Workshop on Collaborative, Open, and Decentralized Training of Foundation Models, selected as an oral presentation. In this work, we introduce a novel decentralized LLM fine-tuning framework that achieves extreme communiation efficiency and global consensus.
Sep 2026 🎓 Our paper on asynchronous pipeline parallelism has been accepted to NeurIPS 2026. In this work, we introduce FRESCO, a new principle for consistency control in asynchronous pipelines, which consistently outperforms standard baselines while being extremely efficient and theoretically sound.
Sep 2026 🎓 Namhoon is serving as an area chair for ICLR 2027.
Aug 2026 🏆 Hyunji, Hyeondo, and Jinseok have been selected for the NRF Research Grant for Master’s and PhD Students.
Aug 2026 🏛 We are co-organizing the LeT Workshop on October 7~8 at POSTECH. The workshop aims to bring together researchers interested in machine learning theory.
Aug 2026 🎓 Our paper on extreme LLM sparsity has been accepted to EMNLP 2026. In this work, we show that carefully handling the elementary components of pruning can already substantially outperform existing pruning methods, even at extreme sparsity.
Aug 2026 🎓 Our paper on semiconductor anomaly detection has been accepted to BMVC 2026. In this work, we introduce a semiconductor defect dataset (OASIS) along with vision-guided instruction tuning method.
Jul 2026 🤝 Our lab has been selected for the 2026 NRF Next-Generation AI for Mathematics program.
Jul 2026 🏆 Kwanhee received the IITP President’s Award in the Graduate Student Outstanding Research Achievement Competition at AI Connect Korea 2026.
Jul 2026 🏆 Our new paper on extreme LLM sparsity has received the Best Paper Award at CKAIA 2026. The preprint will be available on arXiv soon.
Jul 2026 🏆 Sungbin has been selected for the 2026 PhD Excellence Scholarship in Science and Engineering.
Jul 2026 🤝 Our lab joined the National AI Research Lab (NAIRL) academic consortium on July 1, 2026.
Jun 2026 🎓 Our paper on flatness-based out-of-distribution detection has been accepted to ECCV 2026. In this work, we propose FOLD, a lightweight detector that leverages feature-Hessian curvature and partial feature normalization to improve ID-OOD separability.
May 2026 🏆 Our paper on uncertainty quantification in ICL, accepted to ACL 2026, has been selected as an Oral presentation. In this work, we introduce self-function vectors to directly decompose uncertainty, and propose a novel framework to evaluate these disentangled sources.
May 2026 🎓 Our paper on asynchronous pipeline parallelism has been accepted to ICML 2026. In this work, we propose a basis rotation approach to effectively address the associated gradient staleness problem.
Apr 2026 🤝 Our lab has been selected for the IITP Efficient AI National R&D Program (2026–2029), a national research program with KRW 7.5B in total funding, together with leading partners including SqueezeBits.
Feb 2026 🤝 Our lab has been selected for the "Early-Career Researcher Infrastructure Support Program" (NRF) — one of the most competitive grants in Korea. With 500M KRW in funding, we will build a high-performance compute infrastructure to power our research on extreme compression of hyperscale AI foundation models.
Feb 2026 ✈️ Kwanhee started as a visiting student researcher in the Dan Alistarh Group at ISTA.
Jan 2026 🎓 Our paper on extreme LLM sparsity (ELSA) has been accepted to ICLR 2026.