Ye Chen

Ye Chen

An Architecture & Computer Science Student

About Me

Hi! I am Ye Chen (陈晔), a senior undergraduate student at Xi'an Jiaotong University (XJTU) and Politecnico di Milano (POLIMI). Currently, I am pursuing a triple Bachelor degree, including a B.Eng in Computer Science and Technology at XJTU (supervised by Prof. Qin Xia) and a Dual B.Arch in Architecture at XJTU and POLIMI (supervised by Prof. Shanyao Zhu).

It is a great honor that I will join the State Key Laboratory of Blockchain and Data Security in the College of Computer Science and Technology and the School of Cyber Science and Technology at Zhejiang University in 2027-Fall to pursue a Ph.D. in Cyber Science and Technology (supervised by Prof. Kui Ren and Prof. Zhan Qin).

I am a research beginner, full of passion and self-motivated. My research interest includes Trustworthy AI, Multimodal LLM, and Agentic System. In addition, I possess a strong curiosity regarding cutting-edge research topics and interdisciplinary applications.

Currently, I am experiencing my internship as Machine Learning Engineer at OriginArkAI, supervised by Wenhui Dong.

Beside those, I am a keen enthusiast of basketball and swimming. If you would like to connect with me, please feel free to drop me an email or add my WeChat.

🔥 News

  • 2026.09: The work I participated in (CoSec) was announced on Arxiv!
  • 2026.09:  🎉🎉 The work I participated in (POES) was accept by NeurIPS 2026! Find it in Arxiv!
  • 2026.09: The work I participated in (FGPO) was announced on Arxiv!
  • 2026.08: The work I participated in (Emoupdate) was announced on Arxiv!
  • 2026.08:  🎉🎉 The work I participated in (SEPO) was accept by EMNLP 2026 Findings! Find it in Arxiv!
  • 2026.08: The work I participated in (HN-Clip) was announced on Arxiv!
  • 2026.07: The work I participated in (TARA) was announced on Arxiv!
  • 2026.07: The work I participated in (CMVF) was announced on Arxiv!
  • 2026.04:  🎉🎉 The first work I participated in (POES) was announced on Arxiv!

📝 Publications

NeurIPS 2026
sym

Select Smarter, Not More: Prompt-Aware Evaluation Scheduling with Submodular Guarantees

X Ma, Y Li, H Liu, Z Wang, Y Chen, Y Guo, X Tang*

Project

  • A framework that treats APO as an online adaptive testing problem to dynamically select the most informative evaluation subsets for higher accuracy and less computational costs.
EMNLP 2026 Findings
sym

SEPO: Evidence-Grounded Prompt Optimization via Structural Editing

X Ma, H Liu, Y li, J Zhu, Z Wang, Y Chen, X Tang*

Project

  • SEPO is a framework replacing opaque whole-prompt rewrites with structured, evidence-grounded local edits that are traceable, more accurate, and substantially more token-efficient.

📝 Preprints

Arxiv
sym

CoSec: Benchmarking Agent Security in Communities

H Chen, W Dong, Y Chen, J Yao, C Xia, Y Qu, R Wang, F Yuan, C Hamami, C Pan, X Yue, Z Wang, F Ye, C Si, C Shan

Project

  • CoSec is an executable benchmark for evaluating privacy and authorization enforcement in LLM agent systems operating within and across communities.
Arxiv
sym

Why Sample What You Can Enumerate? Exact Policy Optimization for Genomic Tool Selection

H Liu, X Ma, Y Chen, Z Wang, X Tang*

Project

  • FGPO is a method which scores every tool subset and optimizes the exact action expectation, and precomputes the reward of each question–subset pair into an exhaustive table, removing reasoner calls from training.
Arxiv
sym

Do SpeechLMs Hear Their Own Opinions? Diagnosing and Mitigating Previous-Belief Contamination in Streaming Emotion Understanding

H Liu, Z Wang, Y Chen, H Deng, X Tang*

Project

  • Emoupdate is a training-free framework that separates current-audio perception from historical state revision through three components: prior-blind acoustic firewall, evidence-shrunk causal belief filter and decontamination operator.
Arxiv
sym

Which Negatives Matter? Ask Your Text Encoder: Adaptive Similarity Margins for Dense-Caption Retrieval

H Liu, Y Chen, Z Wang, X Tang*

Project

  • HN-CLIP is a method that uses the text encoder’s own text-text geometry to construct per-negative adaptive similarity margins.
Arxiv
sym

Are Prompt Optimizers Blind? Cross-Modal Visual Feedback for Automatic Prompt Optimization

H Liu, X Ma, Y Chen, Y Zou, X Tang*

Project

  • A framework that lets prompt optimizers inspect failed images during optimization, distill recurring visual blind spots into a reusable text prompt, and improve multimodal performance with no additional inference-time cost.
Arxiv
sym

One Rewrite to Fix Them All? Type-Aware Repair Allocation for Text-to-Image Prompt Optimization

H Liu, X Ma, Y Chen, S Cui, X Tang*

Project

  • A training-free framework that diagnoses different text-to-image generation failures, routes each to a type-specific repair, and compiles them into a single optimized prompt to improve semantic fidelity with only one additional regeneration.

🎖 Honors and Awards

  • 2026.05 The Mathematical Contest in Modeling Honorable Mentioned
  • 2025.11 School Scholarship of XJTU-POLIMI Joint School
  • 2025.10 Outstanding Peer Instructor of Xi’an Jiaotong University
  • 2024.12 National Scholarship
  • 2024.05 National TOP 30 of Taobao University Student Innovation Challenge
  • 2023.12 University Scholarship of Xi’an Jiaotong University

📖 Educations

  • 2027.09 - 2032.06 (Expeted), Graduate, Ph.D. in Cyber Science and Technology, Zhejiang University.
  • 2024.03 - 2027.06 (Expeted), Undergraduate, B.Eng in Computer Science and Technology, Xi’an Jiaotong University.
  • 2022.09 - 2027.06 (Expeted), Undergraduate, B.Arch in Architecture, Politecnico di Milano.
  • 2022.09 - 2027.06 (Expeted), Undergraduate, B.Arch in Architecture, Xi’an Jiaotong University.
  • 2019.09 - 2022.06, Zhejiang Ouhai Middle School, Wenzhou.

💻 Internships