Zeqi Lin
Principal Researcher
About
Dr. Zeqi Lin is currently a Principal Researcher at Microsoft CoreAI Post-Training Team. He obtained his Ph.D from Software Engineering Institute, Peking University in 2019 (supervised by Prof. Bing Xie and Prof. Lu Zhang). He joined Microsoft Research Asia (Beijing) in 2019 and Microsoft GenAI (Redmond) in 2024, focusing on LLM-Powered Coding and Reasoning.
Projects
- Reinforcement Learning of GPT-4o Copilot Completion Model
- Phi-3: A Highly Capable Language Model Locally on Your Phone
- Azure OpenAI on Your Data
- Project Sophia, A New Generation AI-First Business Application
- Power BI Quick Measures Using Natural Language
- Power Automate Flow with Natural Language
- Write Power Fx Formulas with Natural Language
- Natural Language Queries in Excel Ideas
- NuGetSolver: A Powerful Tool for Resolving NuGet Dependency Conflicts in Visual Studio
Publications
- Self-Evolved Reward Learning for LLMs (ICLR 2025)
- Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs
- STAND-Guard: A Small Task-Adaptive Content Moderation Model (COLING 2025 Industry Track)
- Make Your LLM Fully Utilize the Context (NeurIPS 2024)
- Can LLMs Learn From Mistakes? An Empirical Study on Reasoning Tasks (EMNLP 2024 Findings)
- Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
- Compositional API Recommendation for Library-Oriented Code Generation (ICPC 2024)
- CodeT: Code Generation with Generated Tests (ICLR 2023)
- Making Language Models Better Reasoners with Step-Aware Verifier (ACL 2023)
- How Do In-Context Examples Affect Compositional Generalization? (ACL 2023)
- Does Deep Learning Learn to Abstract? A Systematic Probing Framework (ICLR 2023)
- Skill-Based Few-Shot Selection for In-Context Learning (EMNLP 2023)
- TAPEX: Table Pre-Training via Learning a Neural SQL Executor (ICLR 2022)
- Nufix: Escape From NuGet Dependency Maze (ICSE 2022)
- Reasoning Like Program Executors (EMNLP 2022)
- CERT: Continual Pre-Training on Sketches for Library-Oriented Code Generation (IJCAI 2022)
- When Language Model Meets Private Library (EMNLP 2022 Findings)
- Can Neural Clone Detection Generalize to Unseen Functionalities? (ASE 2021)
- Learning Algebraic Recombination for Compositional Generalization (ACL 2021 Findings)
- Revisiting Iterative Back-Translation from the Perspective of Compositional Generalization (AAAI 2021)
- Iterative Utterance Segmentation for Neural Semantic Parsing (AAAI 2021)
- Compositional Generalization by Learning Analytical Expressions (NeurIPS 2020)
- Hierarchical Poset Decoding for Compositional Generalization in Language (NeurIPS 2020)
- Adaptive Deep Code Search (ICPC 2020)
- CoRA: Decomposing and Describing Tangled Code Changes for Reviewer (ASE 2019)
- Improving Software Text Retrieval Using Conceptual Knowledge in Source Code (ASE 2017)
- Intelligent Development Environment and Software Knowledge Graph (JCST 2017)
- Mining API Usage Examples from Test Code (ICSME 2014)