Asian Agent Research (日韓)
Asian Agent Research — Japan & Korea
Section titled “Asian Agent Research — Japan & Korea”🇯🇵 Japanese Research
Section titled “🇯🇵 Japanese Research”Sakana AI (Tokyo)
Section titled “Sakana AI (Tokyo)”| Work | Year | Key Contribution |
|---|---|---|
| AI Scientist-v2 | 2025 | First fully AI-generated peer-reviewed paper accepted at ICLR workshop; agentic tree search for autonomous scientific discovery |
| CycleQD | 2025 (ICLR) | Evolutionary algorithm for acquiring multiple agentic skills in LLMs via quality diversity + model merging |
| Evolutionary Model Merge | 2024 | Nature-inspired evolution for combining open-source models; published in Nature Machine Intelligence |
| ShinkaEvolve | 2025 | LLM-driven evolutionary framework; SOTA on circle packing, AIME reasoning |
| M2N2 | 2025 (GECCO) | Nature-inspired competition/attraction for model merging niches |
Preferred Networks (PFN)
Section titled “Preferred Networks (PFN)”| Work | Focus |
|---|---|
| Deep Learning + Autonomous Agents | Core AI research, applied to robotics and industrial automation |
University of Tokyo / Kyoto University
Section titled “University of Tokyo / Kyoto University”| Work | Focus |
|---|---|
| LLM agent architectures | Active research on agent reasoning, tool use, multi-agent coordination |
| Multi-agent systems + AI reasoning | Kyoto University long-standing MAS research tradition |
Japanese Survey
Section titled “Japanese Survey”| Work | Year | Contribution |
|---|---|---|
| Agentic RL Survey (日本語) | 2025 | Comprehensive Japanese survey covering 500+ papers on RL for agentic LLM capabilities |
🇰🇷 Korean Research
Section titled “🇰🇷 Korean Research”KAIST (Daejeon)
Section titled “KAIST (Daejeon)”| Work | Year | Key Contribution |
|---|---|---|
| FluidGPT | 2025 (MobiCom) | UI-based autonomous mobile agent for app control; demonstrated at ACM MobiCom |
| AI Semiconductor ecosystem | Ongoing | KAIST + Naver + Intel collaboration on Gaudi optimization |
| Multi-agent learning | Ongoing | Distributed multi-agent systems research |
Naver / Naver Cloud
Section titled “Naver / Naver Cloud”| Work | Focus |
|---|---|
| HyperCLOVA | Large-scale Korean foundation model |
| Agent N | AI agent platform built on HyperCLOVA |
| Enterprise AI agents | Autonomous workflow agents |
Kakao AI
Section titled “Kakao AI”| Work | Focus |
|---|---|
| LLM agents | Autonomous agent development |
| Multi-agent workflows | Enterprise agent orchestration |
LG AI Research
Section titled “LG AI Research”| Work | Focus |
|---|---|
| Foundation models | Large-scale Korean LLM development |
| Agent systems | Autonomous agent architecture research |
Korean-Language Papers
Section titled “Korean-Language Papers”| Work | Year | Contribution |
|---|---|---|
| LangGraph + CrewAI Hybrid (KSCI) | 2025 | Integration of role-based CrewAI with graph-based LangGraph for Korean-language MAS |
| AgentsNet (한국어) | 2025 | Multi-agent coordination benchmark up to 100 agents; self-organization metrics |
| NAMU Agent Platform | 2025 | Enterprise-grade autonomous agent workflow with 5-stage agentic workflow |
| AIOS: LLM Agent OS (한국어 blog) | 2024 | Korean-language summary of LLM-as-OS-kernel concept |
Japanese/Korean at Top English Venues
Section titled “Japanese/Korean at Top English Venues”Papers at NeurIPS, ICML, ICLR, EMNLP with Japanese/Korean first authors:
| Work | Venue | Year | Contribution |
|---|---|---|---|
| RAP | NeurIPS 2024 | 2024 | Multimodal agents with past experience retrieval for planning |
| AVATAR | NeurIPS 2024 | 2024 | Comparator module for automated prompt optimization in tool use |
| HIAGENT | NeurIPS 2024 | 2024 | Subgoal-based memory chunking for long-horizon tasks |
| Optimus-1 | NeurIPS 2024 | 2024 | KG + experience pool for Minecraft agents |
| Tulip Agent | ICML 2024 | 2024 | Extensible tool library with vector search |
| Agent Q | ICLR 2025 | 2025 | MCTS + DPO for web agent training; 340% improvement |
| ADAS | ICLR 2025 | 2025 | Meta agent search for automatic agent architecture discovery |
| AGENTFLOW | ICLR 2025 | 2025 | Trainable in-the-flow agentic framework with Flow-GRPO |
| Agentic Self-Learning | ICLR 2025 | 2025 | Multi-role closed-loop RL without human supervision |
Key Institutions
Section titled “Key Institutions”| Country | Institution | Specialization |
|---|---|---|
| 🇯🇵 | Sakana AI | Evolutionary AI, agentic systems, model merging |
| 🇯🇵 | Preferred Networks | Deep learning, autonomous agents |
| 🇯🇵 | University of Tokyo | LLM agent architectures |
| 🇯🇵 | Kyoto University | Multi-agent systems, AI reasoning |
| 🇰🇷 | KAIST | Mobile agents, AI semiconductors, multi-agent learning |
| 🇰🇷 | Naver | HyperCLOVA, Agent N, enterprise agents |
| 🇰🇷 | Kakao AI | LLM agents, autonomous workflows |
| 🇰🇷 | LG AI Research | Foundation models, agent systems |