Xingbo Du

Postdoctoral Researcher · MBZUAI

Xingbo Du

I build harnesses that turn language models into dependable problem-solvers for complex tasks, where every answer has to pass an external check.

I am a postdoctoral researcher at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), working with Prof. Le Song on LLM agents for science. I received my Ph.D. in Computer Science from a joint program between Shanghai Jiao Tong University and the University of New South Wales, advised by Prof. Junchi Yan and Prof. Wenjie Zhang.

My work sits between machine learning and optimization. Today I focus on the machinery around the model: grounding requirements as executable constraints, searching over programs and configurations, verifying in stages when evaluation is expensive, and keeping a memory of what worked. Before that, I built learned solvers for combinatorial problems in routing and floorplanning.

Xingbo Du giving a talk
Abu Dhabi, UAE
Oral talk, IJCAI 2019 · Macao
Primary direction

Agent harnesses for complex, verifiable tasks

A strong model is not enough when requirements change mid-task, tools are many, and each evaluation costs hours. The harness around the model decides what it sees, what it may try, how results are checked, and what it remembers. I study each of these parts separately, with fixed budgets and graders the agent cannot edit.

  1. 01

    Grounding

    Turn natural-language requirements into executable, checkable constraints.

  2. 02

    Search

    Search over solver code, parameters, and tool scripts, not just final answers.

  3. 03

    Staged verification

    Run cheap checks first; spend expensive runs only on candidates that survive.

  4. 04

    Memory

    Track what worked, where, and why, and find out what evidence is still missing.

MemR³ NeurIPS'26BioXArena NeurIPS ED'26RSMeM ACL'26LLM4EDA survey
Foundation

Learned solvers for combinatorial optimization

Put problem structure (connectivity, obstacles, module shapes) into the representation, so feasibility holds by construction instead of being repaired afterwards. Applied to routing, Steiner trees, floorplanning, and graph alignment.

HubRouter NeurIPS'23OAREST NeurIPS'25JigsawPlanner ICCAD'24CENALP IJCAI'19 · TKDE'22
Also

Interpretable learning

Concept bottleneck models whose predictions pass through human-readable concepts, and stay accurate when concepts are removed, corrected, or swapped at test time.

DCBM TPAMI'25FCBM AAAI'26
2026
  • NeurIPS 2026MemR³

    Knowing What is Missing: Efficient Conversational Memory via Explicit Evidence-Gap Tracking

    Du Xingbo, Li Loka, Zhang Duzhen, Song Leonard, Song Le

  • NeurIPS ED 2026BioXArena

    BioXArena: Benchmarking LLM Agents on Multi-Modal Biomedical Machine Learning Tasks

    Li Loka, Zhang Duzhen, Du Xingbo, Song Leonard, Wang Zixiao, Aukenov Assanali, Thomas Noel, Sailaukan Shakhnazar, Yang Yonghan, Chen Feilong, Dong Jiahua, Zhang Kun, Zhang Bin, Song Le

  • ICML 2026RulePlanner

    RulePlanner: All-in-One Reinforcement Learner for Unifying Design Rules in 3D Floorplanning

    Zhong Ruizhe, Du Xingbo, Yan Junchi

  • ACL 2026RSMeM

    RSMeM: Knowledge-Enhanced Memory Evolution for Remote Sensing Agents with Systematic Evaluation

    Wu Bingxian*, Zhang Yu*, Guo Zonghao, Liu Tang, Qian Chen, Lu Yuxiang, Du Xingbo, Li Yanghao, Zhang Yidan, Chen Chi, Yao Ling, Sun Maosong

  • AAAI 2026FCBM

    Flexible Concept Bottleneck Model

    Du Xingbo*†, Dou Qiantong*, Fan Lei, Zhang Rui†

2025
  • NeurIPS 2025OAREST

    Train on Pins and Test on Obstacles for Rectilinear Steiner Minimum Tree

    Du Xingbo, Zhong Ruizhe, Yan Junchi

  • IEEE TPAMIDCBM

    The Decoupling Concept Bottleneck Model

    Zhang Rui*, Du Xingbo*, Yan Junchi, Zhang Shihua

  • ICML 2025DSBRouter

    DSBRouter: End-to-end Global Routing via Diffusion Schrödinger Bridge

    Shi Liangliang*, Zhang Shenhui*, Du Xingbo, Yang Nianzu, Yan Junchi

  • ICLR 2025 WorkshopRecurFormer

    RecurFormer: Not All Transformer Heads Need Self-Attention

    Yan Ruiqing*, Zheng Linghan*, Du Xingbo, Zou Han, Guo Yufeng, Yang Jianfei

2024
  • ICCAD 2024JigsawPlanner

    JigsawPlanner: Jigsaw-like Floorplanner for Eliminating Whitespace and Overlap among Complex Rectilinear Modules

    Du Xingbo, Zhong Ruizhe, Kai Shixiong, Tang Zhentao, Xu Siyuan, Hao Jianye, Yuan Mingxuan, Yan Junchi

  • NeurIPS 2024FlexPlanner

    FlexPlanner: Flexible 3D Floorplanning via Deep Reinforcement Learning in Hybrid Action Space with Multi-Modality Representation

    Zhong Ruizhe, Du Xingbo, Kai Shixiong, Tang Zhentao, Xu Siyuan, Hao Jianye, Yuan Mingxuan, Yan Junchi

  • arXivSurvey

    LLM4EDA: Emerging Progress in Large Language Models for Electronic Design Automation

    Zhong Ruizhe*, Du Xingbo*, Kai Shixiong, Tang Zhentao, Xu Siyuan, Zhen Hui-Ling, Hao Jianye, Xu Qiang, Yuan Mingxuan, Yan Junchi

2023
  • NeurIPS 2023HubRouter

    HubRouter: Learning Global Routing via Hub Generation and Pin-hub Connection

    Du Xingbo, Wang Chonghua, Zhong Ruizhe, Yan Junchi

2022
  • IEEE TKDECENALP

    Cross-network Skip-gram Embedding for Joint Network Alignment and Link Prediction

    Du Xingbo, Yan Junchi, Zhang Rui, Zha Hongyuan

2019
  • IJCAI 2019Oral

    Joint Link Prediction and Network Alignment via Cross-graph Embedding

    Du Xingbo, Yan Junchi, Zha Hongyuan

Patents

18 Chinese invention patents · 11 as first inventor · 11 granted
  • 知识图谱构建方法、装置、电子设备及存储介质GrantedCN110807103B · 2026
  • 一种基于线路生成和顺序选择的强化学习全局布线方法GrantedCN116402012B · 2025
  • 知识图谱的噪声识别方法、装置及计算机可读存储介质GrantedCN114048322B · 2025
  • 生成对象简称的方法和系统、存储介质以及程序产品GrantedCN114282545B · 2024
  • 基于知识图谱的推理方法、装置及计算机可读存储介质GrantedCN112989127B · 2024
  • 基于时序关联网络的分析方法、装置及计算机可读存储介质GrantedCN115086144B · 2024
  • 交易检测方法、模型的训练方法、装置及计算机可读存储介质GrantedCN113011979B · 2024
  • 基于知识图谱的异常结构挖掘方法、装置及可读存储介质GrantedCN113704326B · 2024
  • 一种协同训练方法、装置及计算机可读存储介质GrantedCN113705662B · 2024
  • 异常节点识别方法、模型的训练方法、装置及存储介质GrantedCN113362157B · 2024
  • 一种确定对象特征的方法及装置GrantedCN112380494B · 2023
  • 考虑障碍物回避的直角斯坦纳树构建方法和系统CN117390449A · 2024
  • 基于二阶段学习路径规划的芯片全局布线实现方法CN116956811A · 2023
  • 生成式可解释的路径规划方法及设备CN116878530A · 2023
  • 基于改进奖励函数的芯片布局预训练及优化方法CN116362123A · 2023
  • 一种异常卡识别方法、装置、电子设备及存储介质CN114881783A · 2022
  • 对序列数据进行特征处理的方法、装置及存储介质CN113239990A · 2021
  • 无标签对象的分类和检测套码的方法、装置及计算机可读存储介质CN110781955A · 2020
  • BioXArena accepted at NeurIPS ED 2026.
  • MemR³ accepted at NeurIPS 2026.
  • Serving as Area Chair for ICLR 2027.
  • Serving on the Program Committee for AAAI 2027.
  • RulePlanner accepted at ICML 2026.
  • RSMeM accepted at ACL 2026.
  • Serving as reviewer for NeurIPS 2026.
  • Serving as reviewer for ECCV 2026.
  • Serving as reviewer for ACL 2026.
  • Serving as reviewer for ICML 2026.
  • Serving as reviewer for CVPR 2026.
  • FCBM accepted at AAAI 2026.
  • Received my Ph.D. from UNSW.
  • Joined MBZUAI as a postdoctoral researcher.
  • Received my Ph.D. from SJTU.
  • OAREST accepted at NeurIPS 2025.
  • Serving as reviewer for ICLR 2026.
  • Serving on the Program Committee for AAAI 2026.
  • Serving as reviewer for IEEE TCAD.
  • DSBRouter accepted at ICML 2025.
  • RecurFormer accepted at the ICLR 2025 SCOPE Workshop.
  • Serving as reviewer for NeurIPS 2025.
  • Serving as reviewer for ICML 2025.
  • Tutoring COMP9517 Computer Vision at UNSW in 25T1.
  • Invited one-hour talk at the SJTU CSE Global Luncheon.
  • DCBM accepted at IEEE TPAMI.
  • Received the National Scholarship at Shanghai Jiao Tong University.
  • FlexPlanner accepted at NeurIPS 2024.
  • Serving as reviewer for ICLR 2025.
  • Serving as reviewer for IEEE TPAMI.
  • Tutoring COMP9517 Computer Vision at UNSW in 24T3.
  • Serving on the Program Committee for AAAI 2025.
  • JigsawPlanner accepted at ICCAD 2024.
  • Serving as reviewer for NeurIPS 2024.
  • Tutoring COMP9517 Computer Vision at UNSW in 24T2.
  • Our LLM4EDA survey is on arXiv.
  1. 2025 — now

    Postdoctoral Researcher

    Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)
    LLM agents for science · with Prof. Le Song
  2. 2022 — 2025

    Ph.D. in Computer Science

    Shanghai Jiao Tong University & University of New South Wales (joint program)
    Advisors: Prof. Junchi Yan (SJTU) and Prof. Wenjie Zhang (UNSW) · Wu Wenjun Honorary Doctoral Class in AI
  3. 2020 — 2022

    Algorithm Engineer

    China UnionPay, Institute of Financial Technology
    Large-scale graph learning for risk control; member of a National Key R&D Program project on financial fraud detection
  4. 2017 — 2020

    M.Eng. in Software Engineering

    East China Normal University
    Advisor: Prof. Hongyuan Zha
  5. 2013 — 2017

    B.S. in Mathematics

    East China Normal University

Service

  • Area Chair, ICLR 2027
  • Conference reviewer
    ICMLNeurIPSICLRCVPRACLEMNLPECCVAAAICIKM
  • Journal reviewer
    IEEE TPAMIIEEE TCADIEEE TNSEPattern Recognition

Honors

  • Wu Wenjun Honorary Doctoral Scholarship 2023 · 2024
  • National Scholarship 2019 · 2024
  • Shanghai Outstanding Graduate 2017 · 2020

Teaching

  • Tutor, COMP9517 Computer Vision, UNSW 24T2 · 24T3 · 25T1, about 50–60 students per term

Talks & Open Source