RESEARCH · ENGINEERING · LEADERSHIP

Lin Sun

Lin Sun is a senior algorithm researcher and foundation-model team lead in 360’s Zhibrain Division. He also chairs the Algorithm Subcommittee of the 360 Technology Committee, serves on the Academic Committee of the PKU–360 Joint Laboratory for Large Models, and directs the Large-Model Committee at the National Engineering Research Center for Collaborative Security of Big Data. His research focuses on LLM post-training, safety and trustworthy generation, RAG and document intelligence, and memory and experience mechanisms for agents.

A graduate of Harbin Institute of Technology, he has worked in AI R&D since 2010 at Tencent, 360, Beike, and Rushi. As a center general manager at Beike, he led a cross-functional team of around one hundred people and delivered products including Xiaobei Assistant and AI House Tour. The project “Key Technologies and Applications of Large-Model Safety” received the 2025 Beijing Science and Technology Progress Award, Second Prize; Lin Sun participated as the second project contributor. He was selected for the Chaoyang Phoenix Talent program and named a Capital Cybersecurity Talent.

He has published multiple first-author or corresponding-author papers at international conferences including NeurIPS, ACL, EMNLP, and AAAI, and serves as a conference reviewer. He led the training and release of the open reasoning model TinyR1, holds multiple granted invention patents, and contributed to three national and industry standards.

01 / Research

Research & engineering

Foundation models & post-training

Research on long-context data selection, reasoning-path supervision, and model merging to improve reasoning efficiency and generalization.

Safety & trustworthy generation

Research on dynamic attacks, safety alignment, and hierarchical policy control, balancing risk protection, task utility, and generation reliability.

RAG, memory & agents

Research on evidence loss in complex documents, multi-document evidence allocation, long-term memory, and experience transfer in realistic agent workflows.

02 / Selected work

Selected work

2025–2026

TinyR1: reasoning, safety & model merging

Open 32B models spanning branch-merge distillation, safety control, and distribution-aware model merging for reasoning, instruction following, and safety.

2024–2026

LLM safety & trustworthy generation

Research and engineering across dynamic attack–defense, closed-loop training, and detection–response separation; the related project received the Beijing Science and Technology Progress Award, Second Prize.

2024–2026

RAG & trustworthy knowledge services

Document understanding, retrieval, and trustworthy generation, from industrial OCR robustness to budgeted multi-document evidence allocation.

2025–2026

Memory, experience & agent systems

Advances Mem360 long-term memory and experience reuse, with research on agent workflows, memory safety, and experience serving.

03 / Publications

Selected publications

All publications ↗