👋 About Me

I received my bachelor’s degree from Tsinghua University, where I am now a fifth-year Ph.D. student advised by Prof. Yuantao Gu. My research focuses on Agentic Post-Training and Search/Cowork Agents.

I am seeking new graduate roles in LLM post-training. Feel free to reach out about opportunities or collaborations! (WeChat: circleLZY)

🔍 Research Topics

• Agentic Post-Training, Search/Cowork Agent, Change Detection, etc.

🔥 News

• 2026.09: 我们发布了技术报告 “Iris: Climbing to the Search Frontier”，取得同规模开源 Search Agent 的最强综合表现 🚀！ 模型权重与评测 Harness均已开源，欢迎使用与交流！
• 2026.06: 荣获清华大学三好学生.
• 2026.05: 我们的论文 “Stage-wise Distortion-Perception Traversal in Zero-shot Inverse Problems with Diffusion Models” 被 ICML 接收！🎉
• 2026.02: 我们的论文 “JL1-CD: A New Benchmark for Remote Sensing Change Detection and a Robust Multi-Teacher Knowledge Distillation Framework” 被 IEEE TIM 接收！🎉
• 2026.01: 我们的论文 “UnlearningCD: Distill to Forget in Change Detection via Perturbed Knowledge” 被 IEEE JSTARS 接收！🎉
• 2025.11: 荣获第五届《雷达学报》博士论坛最佳报告.
• 2025.10: 荣获《中国图象图形学报》研究生学术沙龙杰出报告.
• 2025.10: 荣获研究生综合奖学金（一等）.
• 2025.09: 我们的论文 “Improving Diffusion-based Inverse Algorithms under Few-Step Constraint via Linear Extrapolation” 被 NeurIPS 2025 接收！🎉
• 2025.06: 荣获清华大学电子之星奖学金.
• 2025.02: 我们的全要素遥感变化检测数据集 JL1-CD 现已开源！ Dataset: JL1-CD Dataset 📂.
• 2025.01: 荣获第二届“吉林一号”杯卫星遥感大赛赛题A（道路提取）冠军，赛题B（变化检测）亚军 🏆.

📖 Educations

• 2022.08-Now, Ph.D. at Department of Electronic Engineering, Tsinghua University.
• 2018.08-2022.06, B.S. at Department of Electronic Engineering, Tsinghua University.

💻 Work Experiences

• 2026.3-2026.09, LLM Intern with Xiaohongshu’s AllSpark-Pevek foundation model team.
• 2025.7-2025.12, LLM Intern at MiroMind.
• 2020.10-2021.10, Research Intern at SenseTime Research.

📝 Selected Publications

My full paper list can be found at .

LLM

Tech Report 2026

Iris: Climbing to the Search Frontier
AllSpark Team (First core contributor)

Project | Weights

• We develop Iris-mini (35B) and Iris-pro (397B).
• Iris achieves the strongest overall results among open-source search agents in the same parameter ranges across BrowseComp, BrowseComp-ZH, DeepSearchQA, and HLE.

Tech Report 2026

MiroFlow: Towards High-Performance and Robust Open-Source Agent Framework for General Deep Research Tasks
MiroMind Team

Project

• We introduce MiroFlow, an open-source framework for general deep research, combining flexible agent graphs, robust workflow execution, and an optional heavy-reasoning mode.
• A unified configuration delivers strong, reproducible results across GAIA, BrowseComp, HLE, xBench-DeepSearch, and FutureX.

Tech Report 2025

MiroThinker: Pushing the Performance Boundaries of Open-Source Research Agents via Model, Context, and Interactive Scaling
MiroMind Team

Project | Weights

• We introduce MiroThinker, an open-source research agent that scales model size, context length, and the depth of agent-environment interactions.
• With a 256K context window and up to 600 tool calls per task, the model sustains long research trajectories and achieves strong results on GAIA, HLE, and BrowseComp.

Remote Sensing Change Detection

TIM 2026

JL1-CD: A New Benchmark for Remote Sensing Change Detection and a Robust Multi-Teacher Knowledge Distillation Framework
Ziyuan Liu, Ruifei Zhu, Long Gao, Yuanxiu Zhou, Jingyu Ma, Yuantao Gu

Project

• We introduce JL1-CD, a new sub-meter, all-inclusive open-source CD dataset comprising 5,000 pairs of remote sensing image patches with a resolution of 0.5–0.75 meters.
• We propose a multi-teacher knowledge distillation (MTKD) framework, which significantly improves the performance of CD models with various network architectures and parameter sizes without increasing any computational or time cost during inference.

JSTARS 2026

UnlearningCD: Distill to Forget in Change Detection via Perturbed Knowledge
Ziyuan Liu, Jiawei Zhang, Qitan Shi, Yuantao Gu

• We propose Unlearning algorithm based on Perturbed Knowledge distillation (UPK), a machine unlearning method applied to CD tasks for the first time.
• The proposed method requires only 10% of the iterations for training the original model and surpasses existing approaches across both closed-box and open-box settings, in both SCD and BDA tasks, and on CNN, Transformer, and Mamba model backbones.

ACM MM 2025

Open-CD: A Comprehensive Toolbox for Change Detection
Kaiyu Li*, Jiawei Jiang*, …, Ziyuan Liu*, Yuantao Gu, …, Xiangyong Cao

Project

• We introduce Open-CD, a unified, modular toolbox for binary and semantic change detection, integrating diverse models, datasets, training pipelines, and analysis tools.
• We benchmark representative methods under a unified evaluation protocol, providing a flexible platform for reproducing existing algorithms and developing new change detectors.

JSTARS 2025

A Novel Multi-Branch Self-Distillation Framework for Optimizing Remote Sensing Change Detection
Ziyuan Liu, Jiawei Zhang, Wenyu Wang, Yuantao Gu

• We propose the MBSD training framework, a novel and generalizable self-distillation learning paradigm designed for CD tasks.
• The proposed framework is single-stage, end-to-end, and incurs only a modest memory and time increase during training, with no extra cost during inference.

GRSL 2025

M$^2$CD: A Unified MultiModal Framework for Optical-SAR Change Detection with Mixture of Experts and Self-Distillation
Ziyuan Liu, Jiawei Zhang, Wenyu Wang, Yuantao Gu

• We propose a unified MultiModal CD framework (M$^2$CD), which is highly versatile and robust, compatible with various backbone architectures.
• By introducing modality-specialized MoE modules into the backbone and innovatively proposing an Optical-to-SAR transition path (O2SP) for self-distillation guidance, we reduce the feature space discrepancies between different modalities and alleviate the model’s burden in processing multimodal data.
• Extensive experiments on the CAU-Flood dataset demonstrate that M2CD outperforms all SOTA methods.

SAR Image Processing

TGRS 2024

DSRKD: Joint Despecking and Super-Resolution of SAR Images via Knowledge Distillation
Ziyuan Liu, Shaoping Wang, Ying Li, Yuantao Gu, Quan Yu

• We propose the DSRKD network, with an encoder–upscaling–decoder architecture that allows for SR at various resolution scales. The encoder and decoder can also be replaced with various more complex modules, and various distillation methods can be easily added, making it a simple and efficient baseline for despeckling SR.
• We innovatively combine the SR of speckled images with denoising using knowledge distillation (KD), significantly improving the performance of the model without introducing any additional computational overhead during inference.

GRSL 2024

SAR Image Compression With Inherent Denoising Capability Through Knowledge Distillation
Ziyuan Liu, Shaoping Wang , Yuantao Gu

• We innovatively utilize a KD mechanism to combine compression with denoising for speckled images. Our method exhibits superior performance compared to other methods on both synthetic and SAR datasets.
• The distillation mechanism is applicable to various learned image compression algorithms and consistently enhances their performance.
• This enhancement is achieved without introducing extra complexity in terms of computational time and memory utilization during inference.

Others

• [Diffusion Model] Manifold-Constrained Noise Optimization for Diverse Diffusion Sampling, Qitan Shi, Cheng Jin, Ziyuan Liu, Yuantao Gu. arXiv 2026.
• [Diffusion Model] Stage-wise Distortion-Perception Traversal in Zero-shot Inverse Problems with Diffusion Models, Jiawei Zhang, Ziyuan Liu, …, Yuantao Gu. ICML 2026.
• [Diffusion Model] Improving Diffusion-based Inverse Algorithms under Few-Step Constraint via Learnable Linear Extrapolation, Jiawei Zhang, Ziyuan Liu, …, Yuantao Gu. NeurIPS 2025.
• [Knowledge Distillation] S3AT: Self-Paced, Self-Distilled, and Self-Finetuned Adversarial Training for Robust Automatic Modulation Recognition, Wenyu Wang, Zhu Lei, Ziyuan Liu, … . TCCN 2025.

🛠 Projects

• Iris | Weights · 核心开发
  Iris-mini、Iris-pro 取得同等参数规模下开源 Search Agent 的最强综合表现，模型权重及完整 Harness 均已开源

• Relax · 参与开发
  小红书开源后训练框架，参与解决模型 MTP 权重转换等问题

• JL1-CD · 独立开发
  提出多教师知识蒸馏框架，提升 20 余种主流变化检测算法的性能；开源完整代码、预训练权重及全要素高分变化检测数据集 JL1-CD

• Open-CD · 核心开发
  基于 OpenMMLab 的综合变化检测工具箱，支持多种模型的训练与评测

• 国家自然科学基金重点项目 · 核心负责
  复杂电磁场景微波图像智能轻量化效能梯度感知技术

🏆 Selected Competitions

• 2025, 第二届“吉林一号”杯卫星遥感应用青年创新创业大赛 赛题A“高分辨率遥感数据道路提取研究” TOP1. (Team leader) (URL)
• 2025, 第二届“吉林一号”杯卫星遥感应用青年创新创业大赛 赛题B“高分辨率遥感影像全要素变化检测研究” TOP2. (Team leader) (URL | Code)

🎖 Honors and Awards

• 2026, 清华大学三好学生.
• 2025, Best Presentation at the 5th Doctoral Forum of the Journal of Radars | 第五届《雷达学报》博士论坛最佳报告.
• 2025, Outstanding Presentation at the Graduate Academic Salon of the Journal of Image and Graphics | 《中国图象图形学报》研究生学术沙龙杰出报告.
• 2025, First-Prize Comprehensive Scholarship for Graduate Students, Tsinghua University | 清华大学研究生综合奖学金（一等）.
• 2025, “Electronic Star” Scholarship of the Department of Electronic Engineering, Tsinghua University | 清华大学电子之星奖学金
• 2025, Champion of Topic A “Road Extraction from High-Resolution Remote Sensing Data” in the 2nd “Jilin-1” Cup Youth Innovation and Entrepreneurship Competition for Satellite Remote Sensing Applications, Jilin Provincial Department of Education | 第二届“吉林一号”杯卫星遥感应用青年创新创业大赛 赛题A“高分辨率遥感数据道路提取研究”冠军，吉林省教育厅
• 2025, Runner-up of Topic B “Full-Element Change Detection in High-Resolution Remote Sensing Images” in the 2nd “Jilin-1” Cup Youth Innovation and Entrepreneurship Competition for Satellite Remote Sensing Applications, Jilin Provincial Department of Education | 第二届“吉林一号”杯卫星遥感应用青年创新创业大赛 赛题B“高分辨率遥感影像全要素变化检测研究”亚军，吉林省教育厅
• 2024, National Scholarship for Graduate Student, Ministry of Education | 研究生国家奖学金
• 2024, First Prize of Lu Dajin & Liu Runsheng Outstanding Teaching Assistant Award, Department of Electronic Engineering, Tsinghua University | 清华大学电子工程系陆大䋮·刘润生优秀助教奖一等奖
• 2021, Sports Scholarship, Tsinghua University | 清华大学体育优秀奖
• 2021, Social Work Scholarship, Tsinghua University | 清华大学社会工作优秀奖
• 2020, Academic Scholarship, Tsinghua University | 清华大学学业优秀奖
• 2020, Sports Scholarship, Tsinghua University | 清华大学体育优秀奖
• 2019, National Scholarship for Undergraduate Student, Ministry of Education | 本科生国家奖学金
• 2019, Comprehensive Scholarship, Tsinghua University | 清华大学综合奖学金
• 2019, Academic Scholarship, Tsinghua University | 清华大学学业优秀奖
• 2019, Sports Scholarship, Tsinghua University | 清华大学体育优秀奖

🏊‍♂️ Sports

• 2026.03: 清华大学马约翰杯游泳比赛甲组男子100米蛙泳冠军、6×50米自由泳接力亚军，助力电子系问鼎游泳马杯甲组亚军 🥈.
• 2025.11: 首届清华大学马约翰杯匹克球比赛冠军 🏆.
• 2025.06: 清华大学游泳研究生联赛50米蛙泳冠军、100米蛙泳冠军、4×50米自由泳接力冠军，助力电子系成就四连冠 🏆.
• 2024.12: 清华大学马约翰杯游泳比赛甲组男子200米蛙泳冠军、100米蛙泳第四、6×50米自由泳接力冠军、4×50米混合泳接力亚军，助力电子系问鼎游泳马杯甲组冠军 🏆.
• 2024.11: 全国城市游泳系列赛青岛站，18-24岁组200米蛙泳冠军、200米仰泳冠军 🏆.
• 2024.09: 第24届全国成人游泳锦标赛（成人游泳爱好者国内最高赛事），18-24岁组200米蛙泳冠军、200米仰泳冠军 🏆.
• 2023.12: 清华大学健美马杯男子模特全场倒数第二 😅.
• 2023.12: 清华大学游泳研究生联赛50米蛙泳冠军、4×50米自由泳接力冠军、4×50米混合泳接力冠军，助力电子系联赛三连冠 🏆.
• 2023.11: 清华大学马约翰杯游泳比赛甲组男子200米蛙泳冠军、50米蛙泳第四 🏆.
• 2023.08: 全国城市游泳系列赛北京站，18-24岁组200米蛙泳亚军、100米蛙泳第四 🏅.
• 2023.06: 宁波梅山湾游泳马拉松挑战赛（全程13.5km）成功完赛.
• 2023.03: 清华大学马约翰杯游泳比赛甲组男子200米蛙泳季军 🏅.
• 2021.11: 清华大学马约翰杯游泳比赛甲组男子200米蛙泳冠军 🏆.
• 2021.03: 清华大学马约翰杯游泳比赛甲组男子200米蛙泳亚军、100米蛙泳第四 🏅.
• 2018.12: 清华大学马约翰杯游泳比赛乙组男子100米蛙泳冠军、50米蛙泳冠军 🏆.

🔗 Academic Service

• Reviewer for NeurIPS, ICML, ICLR, CVPR, ECCV, AAAI, etc.
• Reviewer for TGRS, TITS, TAES, TIM, JSTARS, GRSL, etc.