AI Research
For engineers and researchers who build AI systems
For researchers and students: math foundations, university courses, architectures, training methods, paper feeds, and research labs including Korean academia.
Korean AI Academia
Who is doing frontier AI research in Korea, what their labs work on, and how they are showing up at the top conferences.
Who works on what
Topics in the center, 23 researchers around them, 425 lab members on the outer ring. A lab's slice scales with its size.
Hover or tap a researcher to see their topics, who shares them, and their lab members; tap a topic to see everyone in that area. Connections mean shared research topics, not necessarily co-authorship; outer-ring dots are current grad students from each lab's own page.
Does academia fit the Next Waves?
Each researcher's work mapped onto the five waves of AI value creation — our editorial framework from Next Waves.
The center of gravity is unmistakable: 19 of 23 researchers here work on Wave 2 — frontier models, their evaluation, and their economics. That is the wave Korea already missed commercially and where Chinese open weights are commoditizing value. But look closer and much of this W2 work is really about making models cheap — distillation (DistiLLM-2), KV-cache compression (KVzip), adaptive recursion depth — which is exactly the research a sovereign-AI program running on constrained compute needs.
Wave 3, agentic labor — the wave industry is in right now — has thinner but real coverage (6 researchers): web agents that generalize out of domain, search-reasoning agents, taxonomies for steering how reasoning models think, RLHF and alignment. Wave 4 has genuine seeds (5): simulated patients and hospital workflows winning best-paper awards, fusion-plasma forecasting, brain imaging. And Wave 5 — Korea's structural opening — is where the academic bench lines up best with the industrial thesis (4 labs): humanoid loco-manipulation VLAs, language-driven dexterous grasping, 3D affordance grounding, with roughly 130 grad students in those labs' rosters.
The honest gaps: no one in this AI-faculty set works on Wave 1 — the compute-and-memory wave Korea actually won lives in EE departments and industry fabs, not AI graduate schools. And the W3 gap matters most: industry's current wave is agents shipped as products, and Korea has no academic cluster the size of its model-efficiency cluster working on agentic systems. The verdict: academia is not "off" so much as one wave behind the industrial front — deep where models get cheap, positioned well for embodiment, and thin exactly where the money is moving this year.
KAIST — Kim Jaechul Graduate School of AI
Korea's largest dedicated AI graduate school. Faculty labs span LLMs and generative models, ML theory, NLP, and healthcare AI — and seed a notable share of Korea's AI startups.
At the conferences
The largest single-institution contributor in the NAIRL network's ICML 2026 cohort — 13 papers from 7 PIs — with KAIST labs publishing across NeurIPS, ICML, and ICLR every cycle.
Seoul National University — AI Institute (AIIS)
SNU's interdisciplinary AI institute, connecting labs across engineering, natural sciences, and medicine — deepest in computer vision, deep learning systems, and biointelligence.
At the conferences
SNU vision and ML labs place papers at NeurIPS, ICML, and CVPR every cycle; the Vision & Learning Lab alone reported NeurIPS 2025 and ICML 2025 acceptances.
NAIRL — National AI Research Lab Network
A cross-institution network of AI labs spanning KAIST, Yonsei, Korea University, and POSTECH, pooling frontier research across the member universities.
At the conferences
Announced 31 accepted papers at ICML 2026 across the network — the clearest single number for how Korean university AI research is showing up at top venues.
NAVER AI Lab (TEAM NAVER)
NAVER's research arm behind HyperCLOVA X — publishing on efficient large-scale AI, generative-AI safety, and robotics/physical AI alongside the product work. Former lab head Jung-Woo Ha became Korea's first senior presidential secretary for AI in 2025.
At the conferences
10 papers accepted at NeurIPS 2025 — the strongest disclosed top-venue showing by a Korean corporate lab that year.
LG AI Research
Korea's corporate frontier lab: the EXAONE model family (including open weights), AI for science and materials discovery, and one of the five MSIT sovereign-AI mandates. Former president Kyunghoon Bae became Korea's Minister of Science and ICT.
At the conferences
Publishes EXAONE technical reports and top-venue papers alongside open-weight releases on Hugging Face.
Major Conferences
The venues AI research is measured against — and where Korean labs are showing up.
Neural Information Processing Systems
The field's flagship venue. TEAM NAVER had 10 papers accepted in 2025 — the strongest disclosed showing by a Korean corporate lab.
International Conference on Machine Learning
One of ML's big three. The Korean NAIRL network announced 31 accepted papers in 2026, KAIST the largest single institution at 13.
International Conference on Learning Representations
Deep learning's open-review venue — reviews and rebuttals are public on OpenReview.
Computer Vision and Pattern Recognition
The largest computer-vision venue; a regular target for SNU, KAIST, and NAVER vision labs.
International Conference on Computer Vision
CVPR's odd-year sibling, alternating with ECCV.
European Conference on Computer Vision
Europe's top vision venue, alternating years with ICCV.
Association for Computational Linguistics
NLP's flagship — where Korean-language benchmark and LLM-evaluation work lands.
Empirical Methods in Natural Language Processing
ACL's empirical sibling; strong Korean lab presence in retrieval and evaluation tracks.
AAAI Conference on Artificial Intelligence
The broadest general-AI venue, spanning ML, reasoning, and applications.
Knowledge Discovery and Data Mining
The top venue for data mining and applied ML systems.
Conference on Robot Learning
Where learning meets robotics — the venue to watch as Korea pushes embodied AI.
IEEE International Conference on Robotics and Automation
Robotics' largest venue; relevant to Korea's robot-density and embodiment story.