Research profile · 研究简介

Siyu Lin林思宇

I study intelligent systems that can reason, use tools, and complete long-horizon tasks.

I am a Computer Science and Applied Mathematics student at Yuanpei College, Peking University. My work connects scientific reasoning, post-training, reinforcement learning, and multi-agent systems.

RESEARCH INTERESTS

Scientific Reasoning · Agent Systems · Memory · Reinforcement Learning

Read the research agenda

01 · Research agenda

Research

01Multimodal reasoning · Chemistry · STEMScientific reasoningBuilding evaluation, data, and training systems that help models solve difficult problems grounded in scientific literature and structured knowledge.

02Agent systems · Memory · Task completionLong-horizon agentsStudying how memory, compaction, tool use, and a minimal native harness can make agents more reliable over extended tasks.

03Reinforcement learning · Post-training · Multi-agentScaling post-trainingExploring reinforcement learning, skill-conditioned training, data selection, and multi-agent interaction as routes to stronger models.

Research experience

Where the questions became projects

2026—

Tencent · QingYun Program

Research Engineer Intern

Post-training, scientific agents, and evaluation for complex reasoning systems.

2025

Peking University · HMI Lab

Research Assistant

Reinforcement learning for vision-language models and multimodal benchmark design. Advised by Prof. Shanghang Zhang.

2025

Tsinghua University · NLP Lab

Research Assistant

Long-text processing and interactive survey generation with modular language-model systems. Advised by Dr. Shuo Wang.

2026

UC Berkeley

Research Assistant

Research on intelligent systems and agentic behavior. Advised by Zeyu Zheng.

2025

JoinQuant

Quantitative Research Intern

Alpha research and reinforcement-learning methods for daily factor discovery.

02 · Selected publications

Publications

012025

SUPERChem: A Multimodal Reasoning Benchmark in Chemistry

arXiv preprint · co-first author · under review

022025

SEEA-R1: Tree-Structured Reinforcement Fine-Tuning for Self-Evolving Embodied Agents

NeurIPS 2025 · co-author

032025

LLMxMapReduce-V3: Enabling Interactive In-Depth Survey Generation through a MCP-Driven Hierarchically Modular Agent System

EMNLP 2025 System Demonstration · accepted · co-first author

042025

Loong: Synthesize Long Chain-of-Thoughts at Scale through Verifiers

NeurIPS 2025 Workshop

052026

Economy of Minds: Emerging Multi-Agent Intelligence with Economic Interactions

Under review · co-first author

03 · Visual notes

Field of View

视野之外

Photography is a quieter parallel practice: an archive of attention, light, and the spaces between observations.

Golden morning mist drifting through a wide forest valley — 01Mist, Before MorningStudy 03

Fine reeds silhouetted against a muted evening sky — 02Reeds at DuskStudy 13

An orange chair and table cut by a rectangle of sunlight — 03A Chair for LightStudy 46

A star-filled night sky above a moonlit volcanic valley — 04Altitude, 02:13Study 39

Mountain ridges emerging in cool blue dawn light — 05Blue Hour RidgeStudy 40

Dark mountain silhouettes under a fading amber sky — 06A Line of EveningStudy 31

Orange winter sun reflected across a frozen lake — 07Frozen SunStudy 83

A narrow abstract streak of warm sunset cloud — 08AfterimageStudy 108

A narrow gorge falling toward a green mountain valley — 09Through the GorgeStudy 19

A green mountain valley held between steep ridges — 10Valley, HeldStudy 36

Vertical view into a mountain crater lake — 11Crater Lake IStudy 18-1

Blue crater lake framed by dark volcanic rock — 12Crater Lake IIStudy 18-2

A final vertical study of the crater lake and distant ridges — 13Crater Lake IIIStudy 18-3

Layered mountains receding into silver mist — 14Mountain VeilStudy 76

Snow-covered trees viewed from above — 15White Forest IStudy 78

A solitary person beneath a winter halo — 16White Forest IIStudy 79

A dense forest softened by fresh snow — 17White Forest IIIStudy 80

A pale, minimal sea horizon with distant figures — 18Pale HorizonStudy 15

A lone figure standing on a breakwater beneath an open sky — 19The Long PauseStudy 45

A boat and crescent moon held inside a dark lattice of branches — 20Moon, Boat, BranchStudy 25

A quiet crescent shoreline in pastel light — 21Still CrescentStudy 29

Sea birds circling a soft sunrise — 22First Flight IStudy 94

A flock of birds crossing the morning sea — 23First Flight IIStudy 95

A single bird passing above a low sun — 24First Flight IIIStudy 96

A small lighthouse beside a blue-grey sea — 25Lighthouse in BlueStudy 102

An empty bench facing a lake at dusk — 26Seat at DuskStudy 109

A long band of sunset reflected over calm water — 27Water Keeps the LightStudy 117

A sharp white chapel glowing against a pink dusk sky — 28Chapel at DuskStudy 93

The same chapel at night with two figures below — 29After the BellStudy 97

A coral sunset seen through monumental dark columns — 30Between ColumnsStudy 86

Tree shadows moving across a red wall and grey roof — 31Red Wall, Passing ShadowStudy 112

Pavilion eaves and flowering branches around the sun — 32Eaves in SpringStudy 59

An illuminated pagoda rising between dark branches — 33One Golden TowerStudy 72

A narrow stone passage lit by warm lamps at night — 34Night PassageStudy 120

A snow-covered corner tower reflected in still water — 35White Silence IStudy 88

Red palace walls and white tiled roofs in snow — 36White Silence IIStudy 89

Layered palace roofs receding through snowfall — 37White Silence IIIStudy 91

Rows of chairs crossed by long bands of golden light — 38The Quiet ClassroomStudy 114

A long corridor ending in warm afternoon light — 39Corridor of LightStudy 115

Contact · 联系

For a careful
research question.

I welcome conversations about intelligent systems, scientific reasoning, long-horizon agents, and research collaboration.

siyu_lin@stu.pku.edu.cn ↗Peking University ↗

Beijing · Available by appointment

Siyu Lin林思宇

Research

Where the questions became projects

Research Engineer Intern

Research Assistant

Research Assistant

Research Assistant

Quantitative Research Intern

Publications

SUPERChem: A Multimodal Reasoning Benchmark in Chemistry

SEEA-R1: Tree-Structured Reinforcement Fine-Tuning for Self-Evolving Embodied Agents

LLMxMapReduce-V3: Enabling Interactive In-Depth Survey Generation through a MCP-Driven Hierarchically Modular Agent System

Loong: Synthesize Long Chain-of-Thoughts at Scale through Verifiers

Economy of Minds: Emerging Multi-Agent Intelligence with Economic Interactions

Field of View

First Light

At Altitude

Water Holds Memory

Thresholds

For a careful
research question.