ZJ Zhuohang Jiang

PhD Candidate · Hong Kong Polytechnic University

Zhuohang Jiang

Research Intern @ Kuaishou E-commerce

My research focuses on LLM reasoning, multimodal RAG and self-evolving agents.

I am currently pursuing a PhD degree at the Hong Kong Polytechnic University. My supervisors are Qing Li and Wenqi Fan. My current Cumulative GPA is 3.60/4.00.

I studied at Sichuan University (SCU) from 2020 to 2024, where I majored in Computer Science & Technology. My Major GPA (CS courses): 3.79/4, 89.39/100; Overall GPA: 3.78/4, 89.25/100

During my time at Sichuan University, I worked as a research assistant at MachineILab from 2022 to 2024, advised by Prof. JiZhe Zhou. I participated in one National Natural Science Foundation of China project and one National Key R&D Program of China.

  • 633 Citations
  • 7 h-index
  • Spotlight NeurIPS 2024
  • #12 KDD Cup 2025
NeurIPS 2024Spotlight KDD 2025 KDD 2026ADS Track CVPR 2026Findings AACL-IJCNLP 2026 AAAI 2025 ICONIP 2023 Pattern Recognition

Open to Internship Opportunities

Seeking internships in large language models and AI agents. Feel free to connect via Email or WeChat.
Please note: search, advertising, and recommendation opportunities are not a fit.

01

Research Threads

Six research turns, from tamper detection to agents that ship — and the life that grew around them: scroll to watch each phase unfold.

2023 · 图像篡改检测与定位 Image Forensics
8 baselines

Locating tampered regions patch by patch

tampered authentic
Spotlight NeurIPS 2024

Growing a benchmark, one task at a time

IMDL-BenCo 8 baselines · GPU-accelerated NeurIPS 2024 · Spotlight
39,519 queries

Hierarchical reasoning, from root to leaf

HiBench · 30 tasks 39,519 queries KDD 2025 Benchmark Track
#12 KDD Cup 2025

A web agent's retrieve–read–reason trace

QA-Dragon · dynamic RAG Meta CRAG-MM WebAgents Survey
2,422 egocentric pairs

What the glasses see, and how the agent reasons

SUPERGLASSES SUPERLENS CVPR 2026 Findings
400× AIR throughput

From raw behaviour to online serving

Behaviour raw signals Atomic intents distilled by LLM Online serving retrieval & composition No live LLM call cached atomic knowledge
AIR · KDD'26 ADS Track A/B Agent Kuaishou E-commerce
8 venues reviewed

Teaching, service and teamwork beyond the paper

Teaching

PolyU · 2024–2025

COMP4431 · COMP5423 · COMP2411

AI · NLP Practicum · Database Systems

Academic Service

2023 – Present

TIP · CVPR · ICCV · ECCV · NeurIPS · KDD · AAAI · IoTJ

Reviewer · Frontiers topic coordinator

Leadership & Honours

2022 – 2025

KDD Cup 2025 · team leader · 12th globally

Outstanding Graduate · Tencent Scholarship

PolyU TA · 3 courses Frontiers × 2 journals KDD Cup · #12 globally
Scroll to advance
  1. 2023

    Tamper detection & localisation

    Earlier work at DICALab, Sichuan University focused on computer vision and image forensics; I also attended the NUS Summer School.

    IML-ViTPerceptual MAEICONIP'23
  2. 2024

    From models to benchmarks

    IMDL‑BenCo became a modular benchmark and codebase covering both detection and localisation with 8 baselines and GPU-accelerated evaluation — published in the NeurIPS 2024 Benchmark Track with a Spotlight award. Alongside it, Beyond Visual Appearances used hybrid graph reasoning for privacy-sensitive object identification. Graduated as Outstanding Graduate, then started the PhD at Hong Kong PolyU.

    IMDL-BenCoNeurIPS 2024 SpotlightPrivacy-sensitive Objects
  3. 2025

    Benchmarking hierarchical reasoning

    HiBench, my first-author benchmark, covers 30 hierarchical reasoning tasks with 39,519 queries and was accepted to the KDD 2025 Benchmark Track. The companion survey of WebAgents — towards next-generation AI agents for web automation — was accepted to the KDD 2025 Tutorial Track, and I was appointed Topic Coordinator for two Frontiers journals.

    HiBench · KDD'25WebAgents Survey · KDD'25 Tutorial30 tasks / 39,519 queries
  4. 2025

    Agents that actually retrieve

    The question became whether an agent can finish the job: QA‑Dragon implemented query-aware dynamic RAG for knowledge-intensive visual question answering, choosing the retrieval strategy per query. I led the team at the Meta CRAG‑MM multimodal retrieval challenge and placed 12th globally among hundreds of teams, and gave an invited talk at KDD Reasoning Day on hierarchical data and RAG.

    QA-Dragon · KDD'25 WorkshopMeta CRAG-MM · #12Reasoning Day
  5. 2026

    Agents that see and act

    SUPERGLASSES turned AI smart glasses into an agentic evaluation scenario, testing whether vision-language models can plan and act on a continuous egocentric stream — accepted to CVPR 2026 Findings. I also co-authored the survey on smart glasses for wearable intelligence, and our work on hierarchical pragmatic tactics was accepted to AACL-IJCNLP 2026 Main.

    SUPERGLASSES · CVPR'26AACL-IJCNLP'26 MainSmart Glasses Survey
  6. 2026

    Agents in production

    Developed A/B Agent: aligned historical strategies and code changes, parameters and measured outcomes into retrievable units; integrated automatic experiment comparison, tuning recommendations, configuration checks and review with the A/B platform. Successful and failed experience is continuously written back to support subsequent iterations. Developed AIR: decomposed LLM-generated semantic intent into reusable atomic knowledge. Offline intent generation and caching support online retrieval and composition, approximating sequence-level semantic reasoning without live LLM calls on the serving path.

    AIR · KDD'26 ADSA/B AgentOffline-to-online LLMs
  7. Beyond

    Beyond the paper

    Research is only half the story. Since 2023 I have reviewed for TIP, CVPR, ICCV, ECCV, NeurIPS, KDD, AAAI and IoTJ, and since 2025 I coordinate topics for two Frontiers journals. As a teaching assistant at PolyU I helped run COMP4431 (AI), COMP5423 (NLP Practicum) and COMP2411 (Database Systems); I led the KDD Cup 2025 team to 12th place globally; and I left Sichuan University as an Outstanding Graduate of the university and the province — top-1% GPA, a Tencent Scholarship and a provincial first prize along the way.

    TA · 3 coursesReviewer · 8 venuesFrontiers Topic CoordinatorKDD Cup · #12Outstanding Graduate
02

Research Topics

My research focuses on LLM reasoning and evaluation, self-evolving agents, and multimodal retrieval-augmented generation. I lead work on hierarchical reasoning benchmarks (HiBench), multimodal retrieval agents (SUPERGLASSES / SUPERLENS and QA-Dragon), experience-driven strategy iteration (A/B Agent), and the decoupling of offline LLM intent reasoning from online retrieval (AIR). My earlier research in computer vision is now a supporting background to these directions.

Large Language Models

Reasoning, evaluation and reliable generation with foundation models.

AI Agents

Tool-using agents and self-evolving loops for real-world tasks.

Retrieval-Augmented Generation

Structure-aware RAG for knowledge-intensive and multimodal queries.

Evaluation & Benchmarks

Reasoning and multimodal benchmarks: HiBench and SUPERGLASSES.

03

News

04

Selected Publications

KDD'25 · Benchmark & Dataset Track CCF-A CORE A* (KDD) arXiv GitHub Hugging Face

HiBench: Benchmarking LLMs Capability on Hierarchical Structure Reasoning

Zhuohang Jiang, Pangjing Wu, Ziran Liang, Peter Q. Chen, Xu Yuan, Ye Jia, Jiancheng Tu, Chen Li, Peter H.F. Ng, Qing Li

Hierarchical ReasoningLLM Benchmark30 Tasks39,519 Queries

As first author and project lead, I led benchmark design and code development for evaluating hierarchical structure reasoning in LLMs. I built tree, JSON and formula generators with controlled depth, width and node counts, algorithmic ground-truth answers and unified prompts, and incorporated code and paper tasks. The benchmark contains 30 tasks and 39,519 queries.

We evaluated 20 LLMs from 10 families, comparing structural representations, complexity and chain-of-thought prompting. I constructed instruction-tuning data that yielded average relative gains of 88.84% for Llama-3.1-8B and 31.38% for Qwen2.5-7B. The dataset and modular evaluation toolkit are open-source, supporting reproducible evaluation and further research on structure-aware reasoning.

CVPR'26 Findings CCF-A arXiv

SUPERGLASSES: Benchmarking Vision Language Models as Intelligent Agents for AI Smart Glasses

Zhuohang Jiang, Xu Yuan, Haohao Qu, Shanru Lin, Kanglong Liu, Wenqi Fan, Qing Li

AI Smart GlassesVQA BenchmarkVision-Language ModelsMultimodal RAG

As first author, I built SUPERGLASSES, a benchmark for evaluating vision-language models as smart-glasses agents, and SUPERLENS, the accompanying multimodal agent. The real-device dataset contains 2,422 QA pairs across 14 domains and 8 question types, with tool-use traces. Domain routing selects direct answering or RAG, while open-vocabulary detection localizes small objects and multi-hop questions are decomposed into single-hop subqueries.

The agent searches image and text evidence, cleans HTML and reranks evidence chunks. A two-level Redis cache reuses query-to-URL and URL-to-content results. We evaluated 26 model/RAG variants; SUPERLENS improved the Qwen2.5-VL-7B score from 32.82 to 44.10, a gain of 11.28 points.

arXiv Preprint arXiv

A/B Agent: A Self-Evolving Agent for Strategy Iteration in Industrial A/B Testing

Zhuohang Jiang, Yuxin Chen, Yongsen Pan, Zheng Hu, Wenqi Fan, Qing Li, Hongyang Wang, Jun Wang, Wenwu Ou

Self-Evolving AgentsTree-RAGA/B TestingStrategy Optimization

As first author, I developed a self-evolving LLM agent for strategy iteration in industrial A/B testing. It organizes experience by business domain, scenario, stage and optimization objective. Hybrid sparse/dense retrieval, tree-path proximity weighting and Qwen-Reranker retrieve relevant evidence; generated strategy configurations are checked for engineering feasibility and guardrail risks.

The experiment tree records strategies, parameters, metrics and context, enabling parent-child and sibling-branch comparisons. Confidence-weighted gains and guardrail-loss penalties guide local search within valid parameter ranges, or a switch to another strategy branch when gains stall. Recording successful and failed trials provides a reusable feedback loop for subsequent iterations. This work is a 2026 arXiv preprint.

KDD'25 Workshop CCF-A arXiv

QA‑Dragon: Query‑Aware Dynamic RAG System for Knowledge‑Intensive Visual Question Answering

Zhuohang Jiang, Pangjing Wu, Xu Yuan, Wenqi Fan, Qing Li

Query-Aware RAGKnowledge-Intensive VQAMulti-Hop Reasoning

I am the first author of QA-Dragon, a multimodal retrieval-augmented generation system for complex visual questions. The system drafts a candidate answer first, then decides in a query-aware manner whether that draft should be returned directly, verified through search, or handed to the retrieval-augmented pipeline.

Tool selection follows the detected knowledge gap of the draft, and the retrieval pipeline combines target cropping, image and text retrieval, evidence reranking, and answer verification to ground the final answer in retrieved evidence. Under single-source, multi-source, and multi-turn question settings, the system improves over the baseline by 5.06%, 6.35%, and 5.03% respectively.

KDD'26 · ADS Track CCF-A CORE A* (KDD) arXiv

Atomic Intent Reasoning: Bringing LLM Semantics to Industrial Cross-Domain Recommendations

Zhuohang Jiang, Yuxin Chen, Shijie Wang, Haohao Qu, Zhou Jindong, Wenqi Fan, Li Qing, Dongxu Liang, Jun Wang

LLM ReasoningAtomic IntentsOffline-to-Online RetrievalIndustrial Deployment

As first author, I developed AIR to decouple LLM intent reasoning from online serving. Offline, an LLM maps behavior records, user attributes and object descriptions into hierarchical intent paths, stored as reusable atomic knowledge. This moves semantic generation out of the online request path and makes the resulting knowledge cacheable and composable.

Online, AIR builds intent trees with behavior weights and time decay, retrieves target-relevant intent chains and fuses evidence with multi-head attention. Retrieval and composition approximate sequence-level semantic reasoning without live LLM calls. The system achieves approximately 400× throughput improvement, demonstrating an efficient path from offline LLM-generated knowledge to online evidence retrieval and composition.

KDD'25 CCF-A arXiv

A Survey of WebAgents: Towards Next-Generation AI Agents for Web Automation with Large Foundation Models

Liangbo Ning, Ziran Liang, Zhuohang Jiang, Haohao Qu, Yujuan Ding, Wenqi Fan, Xiao-yong Wei, Shanru Lin, Hui Liu, Philip S. Yu, Qing Li

Web AgentsFoundation ModelsWeb AutomationTrustworthiness

A survey of web automation agents built on large foundation models. This work complements my implementation research on retrieval and tool-using agents.

AACL-IJCNLP'26 Main AACL-IJCNLP 2026 Main CORE B (IJCNLP)

"Do You Truly Love Me?" Benchmarking LLM Capability on Hierarchical Pragmatic Tactic Conversations

Peter Q. Chen, Pangjing Wu, Zhuohang Jiang, Xiaodong Li, Peter H. F. Ng

LLM EvaluationHierarchical PragmaticsConversational Reasoning

Evaluates LLM capability in hierarchical pragmatic tactic conversations, extending my research on structured reasoning evaluation to conversational contexts.

AAAI'25 CCF-A CORE A* (AAAI) arXiv

Mesoscopic Insights: Orchestrating Multi-Scale & Hybrid Architecture for Image Manipulation Localization

Xuekang Zhu, Xiaochen Ma, Lei Su, Zhuohang Jiang, Bo Du, Xiwen Wang, Zeyu Lei, Wentao Feng, Chi-Man Pun, Jizhe Zhou

Image Manipulation LocalizationMulti-Scale FeaturesHybrid Architecture
NeurIPS'24 CCF-A CORE A* (NeurIPS) arXiv GitHub

IMDL-BenCo: A Comprehensive Benchmark and Codebase for Image Manipulation Detection & Localization

Xiaochen Ma, Xuekang Zhu, Lei Su, Bo Du, Zhuohang Jiang, Bingkui Tong, Zeyu Lei, Xinyu Yang, Chi-Man Pun, Jiancheng Lv, Jizhe Zhou

Image ForensicsModular BenchmarkRobustness EvaluationReproducibility
ICONIP'23 CCF-C CORE B (ICONIP) Paper

TPTGAN: Two-Path Transformer-Based Generative Adversarial Network Using Joint Magnitude Masking and Complex Spectral Mapping for Speech Enhancement

Zhaoyi Liu, Zhuohang Jiang, Wendian Luo, Zhuoyao Fan, Haoda Di, Yufan Long, Haizhou Wang

Speech EnhancementTwo-Path TransformerGANSpectral Mapping
Pattern Recognition CCF-B · Pattern Recognition Journal Major Revision

TMSC: A Tri-Regional Mask-Guided Semantic Calibration Framework for Zero-Shot Anomaly Detection

Wendian Luo, Tong Yu, Junnan Li, Shengxin Dai, Bing Guo, and Zhuohang Jiang

Zero-Shot Anomaly DetectionTri-Regional MasksSemantic Calibration
arXiv Preprint arXiv

SIDRec: End-to-End Semantic ID Representation and Modeling for Recommendation

Zhuohang Jiang, Yongsen Pan, Pangjing Wu, Zihuai Zhao, Zheng Hu, Yuxin Chen

Semantic IDsGenerative RecommendationCollaborative FilteringEnd-to-End Optimization
arXiv Preprint arXiv

HiGenTrans: Hierarchical LLM Knowledge Transfer for Generative Recommendation

Zhuohang Jiang, Yuxin Chen, Haohao Qu, ZeFei Luo, Yunlong Wang, Xu Yuan, Haojie Ding, Boheng Zhang, Fan Yang, Hongyang Wang, Jun Wang, Zhongbo Sun, Wenqi Fan, Qing Li, Wenwu Ou, Peng Jiang

LLM Knowledge TransferGenerative RecommendationSemantic IDsMulti-Signal Calibration
arXiv Preprint arXiv

HiPaper: A Human-AI Collaborative System for Academic Literature Triage and Scholarly Sensemaking

Ye Jia, Pangjing Wu, Hayley Y. W. Gu, Zhuohang Jiang, Xu Yuan, Peter H. F. Ng, Zackary P. T. Sin, Qing Li, Chen Li

Human-AI CollaborationLiterature ReviewScholarly SensemakingInformation Retrieval
Preprints.org Preprints.org

A Survey on AI Smart Glasses for Wearable Intelligence: From Egocentric Sensing to Agentic Personalization

Xu Yuan, Yi Wang, Zhuohang Jiang, Haohao Qu, Yujuan Ding, Shanru Lin, Guoliang Xing, Hongxia Yang, Jiannong Cao, Qing Li, Wenqi Fan

AI Smart GlassesEgocentric SensingMultimodal AgentsPersonalization
arXiv Preprint arXiv

Beyond Visual Appearances: Privacy-sensitive Objects Identification via Hybrid Graph Reasoning

Xiaochen Ma, Xuekang Zhu, Lei Su, Bo Du, Zhuohang Jiang, Bingkui Tong, Xia Du, Ahmed Alhammadi, Jizhe Zhou

Privacy-Sensitive ObjectsScene GraphsHybrid Graph Reasoning
arXiv Preprint arXiv

IML-ViT: Benchmarking Image Manipulation Localization by Vision Transformer

Xiaochen Ma, Bo Du, Zhuohang Jiang, Ahmed Y. Al Hammadi, Jizhe Zhou

Image Manipulation LocalizationVision TransformerMulti-Scale FeaturesEdge Supervision
arXiv Preprint arXiv

Perceptual MAE for Image Manipulation Localization: A High-level Vision Learner Focusing on Low-level Features

Xiaochen Ma, Zhuohang Jiang, Xiong Xu, Chi-Man Pun, Jizhe Zhou

Image Manipulation LocalizationMasked AutoencodersPerceptual Supervision
05

Selected Projects

Textus: The Next-Generation Markdown Editor and Knowledge Tool

Project Partner • Open-Source Product • 2026

GitHub GitHub Stars Website

AI WritingMarkdown EditorKnowledge ManagementRust + Tauri580 Stars

Meta CRAG-MM: Multimodal Retrieval Challenge

Team Leader • KDD Cup 2025 • 2025

Multimodal RetrievalRAGTeam LeadershipGlobal Rank #12

IMDL-BenCo: Benchmark for Image Manipulation Detection & Localization

Co-First Author • NeurIPS 2024 Benchmark Track — Spotlight • 2024

arXiv GitHub

Image Forensics8 BaselinesGPU-Accelerated EvaluationNeurIPS 2024 Spotlight
06

Invited Talks & Tutorials

Invited Talk

Understanding Hierarchical Data with Large Language Models: RAG, Structural Reasoning, and Future Directions

Invited Talk • Reasoning Day @ KDD 2025 • Toronto, ON, Canada • Aug 2025

Invited TalkHierarchical ReasoningRAGKDD 2025 Reasoning Day
Tutorial

WebAgents: Towards Next-Generation AI Agents for Web Automation with Large Foundation Models

Yujuan Ding, Liangbo Ning, Ziran Liang, Zhuohang Jiang, Haohao Qu, Wenqi Fan, Qing Li, Hui Liu, Xiaoyong Wei, Philip S. Yu

KDD 2025 Tutorial • The 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2025) • 2025

arXiv

KDD 2025 TutorialWeb AgentsFoundation ModelsWeb Automation
07

Education

Institution Degree Period
Hong Kong Polytechnic University Hongkong, China PhD in Computer Science and Technology Sep. 2024 — Present
Sichuan University Chengdu, Sichuan, China B.E. in Computer Science and Technology Sep. 2020 — Jun. 2024
08

Experience

Position Responsibilities
Kuaishou E-commerce Team Kuaishou Inc.
Research Intern Oct. 2025 — Present · Beijing, China
  • Developed A/B Agent: aligned historical strategies and code changes, parameters and measured outcomes into retrievable units; integrated automatic experiment comparison, tuning recommendations, configuration checks and review with the A/B platform. Successful and failed experience is continuously written back to support subsequent iterations.
  • Developed AIR: decomposed LLM-generated semantic intent into reusable atomic knowledge. Offline intent generation and caching support online retrieval and composition, approximating sequence-level semantic reasoning without live LLM calls on the serving path.
DICALab Sichuan University
Research Assistant Sep. 2022 — Jun. 2024 · Advisor: Prof. JiZhe Zhou
  • Earlier research on computer vision and image forensics, including national research projects and peer-reviewed publications.
National University of Singapore (NUS) School of Computing
Summer School Participant Aug. 2023
  • Participated in the School of Computing summer research programme.
09

Selected Awards

Award Institution & Year Distinction
KDD Cup 2025 — Meta CRAG-MM Toronto, Canada, 2025 12th Place (Global)
NeurIPS 2024 — IMDL‑BenCo (Co‑first Author) Vancouver, Canada, 2024 Spotlight Award
Outstanding Graduate Sichuan University & Sichuan Province, 2024 Top Achievement
Tencent Scholarship Sichuan University, China, 2023 Top 2%
A-Level Certificate Comprehensive Quality Evaluation, China, 2023 Excellence
Comprehensive First Class Scholarship Sichuan University, Sichuan, China, 2022 Top 1%
Outstanding Students of Sichuan University Sichuan University, Sichuan, China, 2022 Top 5%
10

Professional Service

Role Venue & Period Details
Conference & Journal Reviewer 2023 – Present
TIPCVPRICCVECCVNeurIPSKDDAAAIIoTJ
Conference & Journal Topic Coordinator 2025-2026
Frontiers in Artificial IntelligenceFrontiers in Big Data
Teaching Assistant The Hong Kong Polytechnic University (PolyU)
Artificial Intelligence (COMP4431)NLP Practicum (COMP5423)DataBase System (COMP2411)
11

Skills

Category Items
Research Topics
Large Language Models (LLMs)AI AgentsRetrieval‑Augmented Generation (RAG)LLM Reasoning & Evaluation
Frameworks & Tools
PyTorchHugging FaceNumPyDockerGitAnaconda
Languages
Mandarin (native)English (fluent, IELTS 6.5)
12

Friends