Jul 2026 – Present
University of Wisconsin–Madison · Department of Statistics
Research Intern.
01 / UPDATES
News
I received the BYD Scholarship and the Hero Scholarship awarded by Hero Esports.
CASTLE was selected for an oral presentation at EMNLP 2026.
Information Processing & Management accepted EduAgentQG for publication.
PRP was published in IEEE/ACM Transactions on Audio, Speech, and Language Processing.
From Physics to Representation was accepted to ACM ICMR 2026.
TopoSHIELD was accepted to Findings of ACL 2026.
Began research at the Institute of Software, Chinese Academy of Sciences, on physics-inspired neural networks and multimodal domain adaptation.
CASTLE was first released on arXiv.
STAR-RFF was accepted to IEEE ICASSP 2026.
Began research at the Institute of Psychology and Behavioral Science, Shanghai Jiao Tong University, on Sim2Real audio representation learning.
The Emergence of Shared Norms in Decentralized Networks was accepted to ICSC 2025.
SMRC was first released on arXiv.
EduAgentQG was first released on arXiv.
Began research on personalized alignment for AI safety at the Shanghai Institute for Intelligent Education.
02 / RESEARCH
Research Experience
Mar 2026 – Aug 2026
Institute of Software, Chinese Academy of Sciences · State Key Laboratory of Space-Ground Integrated Information Systems
Research on physics-inspired neural networks and multimodal domain adaptation.
Aug 2025 – Aug 2026
Shanghai Institute for Intelligent Education (SiiE)
Research on personalized alignment for AI safety.
Dec 2025 – Present
Institute of Psychology and Behavioral Science, Shanghai Jiao Tong University
Research on audio representation learning for Sim2Real transfer.
03 / OUTPUT
Selected Publications
PRP: Procedural-to-Real Masked Pre-training for Transferable and Interpretable Audio Representations
Introduces procedural-to-real masked pre-training for transferable and interpretable audio representations.
From Physics to Representation: Audio Learning with Synthetic Pre-training via Procedural Generation
Introduces AudioPG, trained without real audio; it reaches 90.60% on ESC-50 after less than 20 minutes of single-GPU pre-training.
STAR-RFF: Spatio-Temporal Sensing Assisted Robust Radio Frequency Fingerprint Identification Via STGCN
Uses spatio-temporal graph modeling to improve robust radio-frequency fingerprint identification under changing sensing conditions.
The Emergence of Shared Norms in Decentralized Networks: A Social Computing Model of Lexical Coordination
Proposes a semantic coordination game to quantify the trade-off between communication efficiency and expressive capacity in scale-free networks.
CASTLE: A Comprehensive Benchmark for Evaluating Student-Tailored Personalized Safety in Large Language Models
Builds a bilingual benchmark of 92,908 examples spanning 15 educational-risk dimensions and 14 student attributes.
SMRC: Aligning Large Language Models with Student Reasoning for Mathematical Error Correction
Uses Monte Carlo Tree Search and process reward models to preserve students' correct reasoning steps during mathematical error correction.
EduAgentQG: A Multi-Agent Workflow Framework for Personalized Question Generation
Introduces a multi-agent workflow for personalized question generation.
TopoSHIELD: Reshaping the Flow of Malice via Spatio-Temporal Risk-Aware Topological Evolution in Multi-Agent Systems
Dynamically reshapes multi-agent interaction topology to suppress the propagation of malicious information.
PREPRINTS
Recent Preprints
Sep 2026
Sample-Conditioned Representation Selection for Audio Few-Shot Learning
SAMPLESELECT learns a sample-conditioned feature mask for few-shot audio classification under background shifts while keeping the encoder frozen.
Sep 2026
Rethinking Procedural Audio Pre-training: Source Scaling and Objective Adaptation
Separates formula-class coverage from rendering diversity and examines how procedural audio sources affect scaling benefits and pre-training objectives.
04 / TRAINING
Education
2024 – Present
East China Normal University (ECNU)
Major-course ranking: top 15% · Comprehensive evaluation rank: 1st · BYD Scholarship, 2026. Project 985, Project 211, Double First-Class University (Class A).
Selected coursework: Introduction to Computer Science 94/100 · Data Visualization 93/100 · Artificial Intelligence 92/100 · Machine Learning 89/100 · Mathematical Analysis II 89/100
2018 – 2024
Haimen High School of Jiangsu Province
Secondary education.
05 / BUILD
Selected Projects
Moral Intelligence Education Platform
Project Lead · 2026 – PresentA multi-agent collaborative cloud platform for personalized moral education in underserved and remote regions.
ProfReach
Independent Developer · 2025 – PresentA local-first academic outreach system combining scholar retrieval, retrieval-augmented generation, personalized drafting, and human-in-the-loop safety controls.
MindFlow
Project Lead · 2025 – PresentA privacy-conscious psychological sandbox system with multimodal data collection and browser-side inference using ONNX Runtime Web.
06 / RECOGNITION
Selected Honors
- 2026BYD Scholarship (Top 0.1% nationwide), RMB 10,000.
- 2026Hero Scholarship, awarded by Hero Esports; RMB 10,000.
- 2026National Second Prize and Regional Grand Prize, National College Student E-Commerce Innovation, Creativity and Entrepreneurship Challenge.
- 2026Second Place in the National Final and First Prize in the track final, Tianyi Cloud Cup Shanghai Cloud Application Competition.
- 2025National Second Prize and Regional First Prize, China Mobile M-Zone AI+ University Innovation Program.
- 2025First Place, ECNU Artificial Intelligence Innovation Application Competition, Agent Track.
- 2024 – 2025ECNU Second-Class Scholarship and Outstanding Student Award.
07 / COMMUNITY
Academic Service & Grants
Academic Service
- International Conference on Learning Representations
- AAAI Conference on Artificial Intelligence
- IEEE/ACM Transactions on Audio, Speech, and Language Processing
Research Grants
- National Undergraduate Innovation Training Program
- Innovation Program, State Key Laboratory of Space-Ground Integrated Information Systems
- ECNU Future Leading Talent Excellence Academic Program