Scholar
Keqing He
Google Scholar ID: 811USNoAAAAJ
Unknown affiliation
Citations & Impact
Citations
1,398
H-index
20
i10-index
37
Publications
20
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- ICLR 2025: "AgentRefine: Enhancing Agent Generalization through Refinement Tuning"
- ACL 2024: "DolphCoder: Echo-Locating Code Large Language Models with Diverse and Multi-Objective Instruction Tuning"
- EMNLP 2024: "How Do Your Code LLMs perform? Empowering Code Instruction Tuning with Really Good Data"
- COLING 2024: 2 papers accepted
- ACL 2024: 1 paper accepted
- ICLR 2023: 1 paper accepted
- EMNLP 2023: 4 papers accepted
- ACL 2023: 4 papers accepted
- EMNLP 2022: 4 papers accepted
- COLING 2022: 3 papers accepted
- CIKM 2022: 1 paper accepted
- NAACL 2022: 2 papers accepted
- SIGIR 2022: 1 paper accepted
- ACL 2022: 1 paper accepted
- arXiv: "SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wild"
Research Experience
- Mar 2023–Present: Full-time Researcher at Meituan LLM Group, focusing on reasoning models, MoE, and LLM alignment
- Jun 2021–Mar 2023: Full-time Researcher at Meituan NLP Group, working on dialogue systems and dialogue pretraining
- Jun 2020–Oct 2020: Research Intern at Alibaba DAMO Academy, focusing on recommendation systems
- Mar 2020–Jun 2020: Research Intern at Tencent WeChat AI Lab, working on zero-shot learning and slot filling
- Oct 2019–Mar 2020: Research Intern at Meituan NLP Group, researching GCN and dialogue systems
Background
- Currently working at Meituan LLM Team, with research focus on reasoning models (e.g., o1), Mixture of Experts (MoE), and LLM alignment.
- Research interests center on three key areas of Large Language Models (LLMs): complex reasoning, reinforcement learning in real-world settings, and LLM alignment.
- In complex reasoning, focuses on the evolution of foundational models and optimization of Long-COT RL, aiming to build new technical pipelines from pre-training to post-training.
- In real-world reinforcement learning, explores LLM-driven end-to-end agent systems (e.g., DeepResearch, GUI Agent, Embodied Agent) to push intelligence boundaries through interaction with dynamic environments.
- In LLM alignment, works on scalable alignment learning, including data evaluation/optimization and preference learning algorithms, to ensure models are both powerful and aligned with human values.
Co-authors