Mingda Zhang

Mingda Zhang

About me

Hello! I’m Mingda Zhang, an undergraduate at the School of Software, Yunnan University (2022.9–2026.6). My research interests include Multimodal Information Fusion, AI Agent & Multi-agent Systems, and AI4Science (SCI, Law, and Bioinformatics), aiming to leverage AI for solving real‑world problems.

I’m fortunate to be mentored by Professor Xiaoyang Tan (NUAA), Professor Jianglong Qin (Software, YNU), and Professor Qing Xu (Law, YNU), whose guidance has helped me develop a systematic, problem‑driven approach to AI research.

🎓 Education and Internships

  • Yunnan University, School of Software — B.Eng. (2022.9 – 2026.6)
  • China Telecom, Chongqing Branch — Technical Engineer (2024.7 – 2024.9)
  • The Chinese University of Hong Kong, Shenzhen — Visiting Student (2025.10 – 2026.4)

🔬 Research Highlights

Agentic Workflow Orchestration

  • Collaborated with Nanyang Technological University (NTU) and National University of Singapore (NUS) on “FlowSteer: Interactive Agentic Workflow Orchestration via End-to-End Reinforcement Learning”, proposing an end-to-end RL framework for automated workflow orchestration through multi-turn agent–environment interaction. Homepage

Medical Image Analysis & Segmentation

  • Author/co‑author on lung disease recognition in collaboration with Army Medical University (AMU), resulting in the paper “A Semantic Segmentation Algorithm for Pleural Effusion Based on DBIF-AUNet” and a patent, which have been deployed in clinical practice.

Applications of Large Language Models

  • Collaborated with the President of an Intermediate People’s Court; our system received multiple letters of recommendation.
  • Sep 2025: National Social Science Fund of China (NSSFC) project officially approved —
    “An Empirical Study on the Mechanism of Judicial Justice in the Digital‑Intelligence Era under Socialism with Chinese Characteristics.” - Number:25CFX009

GitHub — Open Source Projects

Efficient fine‑tuning of LLMs for the medical vertical.
GitHub Stars Repo

FlowSteer — Interactive Agentic Workflow Orchestration via End-to-End RL.
GitHub Stars Repo

💼 Others

  • ICASSP 2026 reviewer
  • University-level scholarship
  • National Gold Award, “New Humanities and Social Sciences Practice and Innovation Competition for College Students”

📚 Papers

  1. “FlowSteer: Interactive Agentic Workflow Orchestration via End-to-End Reinforcement Learning” (ICML 2026) — An end-to-end RL framework for automated workflow orchestration through multi-turn agent–canvas interaction with diversity-constrained rewards. (Mingda Zhang*, Haoran Luo*, Tiesunlong Shen, Qika Lin, Xiaoying Tang, Rui Mao, Erik Cambria) PDF

  2. “Multimodal Fusion at Three Tiers: Physics-Driven Data Generation and Vision-Language Guidance for Brain Tumor Segmentation” (BMC Medical Imaging, JCR Q1) [Accepted] — Bidirectional interaction between VLM models and deep learning for brain tumor recognition. (Mingda Zhang, Kaiwen Pan) PDF

  3. “An Integrated Framework of Prompt Engineering and Multidimensional Knowledge Graphs for Legal Dispute Analysis” (Scientific Reports, JCR Q1) [Accepted] — Combining prompt engineering with knowledge graphs for legal analysis. (Mingda Zhang, Na Zhao, Jianglong Qin, Qing Xu, Kaiwen Pan, Ting Luo) PDF

  4. “Knowledge-Guided Brain Tumor Segmentation via Synchronized Visual-Semantic-Topological Prior Fusion” (Array, JCR Q1) — Integrating VLM models and deep learning for brain tumor recognition. (Mingda Zhang — solo author) PDF

  5. “DCFFSNet: Deep Connectivity Feature Fusion Separation Network for Medical Image Segmentation” (iScience, JCR Q1) — Proposed a feature space decoupling strategy to address segmentation fragmentation caused by forced feature coupling. (Mingda Zhang, Xun Ye, Ruixiang Tang, Jianglong Qin) PDF

  6. “The Consistency-Acceptability Divergence of LLMs in Judicial Decision-Making: Task and Stakeholder Dimensions” (Humanities & Social Sciences Communications, JCR Q1 / CAS Q1) — Proposing the concept of Consistency-Acceptability Divergence to reveal social phenomena in current LLMs. (Mingda Zhang, Xu Qing) PDF

  7. “A Multi-granularity Sparse Concept Activation and Hierarchical Knowledge Graph Fusion Framework for Rare Disease Diagnosis” (iScience, JCR Q1) — Addressing knowledge deficiency and algorithms to LLM in rare disease diagnosis. (Mingda Zhang, Na Zhao, Jianglong Qin, Guoyu Ye, Ruixiang Tang) PDF

  8. “A Method for the Architecture of a Medical Vertical Large Language Model Based on Deepseek R1” (CCF-AI-2025) [Accepted] — Designing a lightweight LLM architecture for the bioinformatics field. (Mingda Zhang, Jianglong Qin) PDF

  9. “ELPG-DTFS: Prior-Guided Adaptive Time-Frequency Graph Neural Network for EEG Depression Diagnosis” (IJCNN 2026) — Modeling EEG dynamic connectivity through channel-band attention, learnable connections and residual prior integration. (Jingru Qiu, Jiale Liang, Xuanhan Fan, Mingda Zhang, Zhenli He) PDF

  10. “A Semantic Segmentation Algorithm for Pleural Effusion Based on DBIF-AUNet” (Computer Technology and Development) [Accepted] — Semantic segmentation of pleural effusion CT via dual-domain feature disentanglement and multi-branch interactive attention. (Ruixiang Tang, Mingda Zhang, Jianglong Qin, Yan Song, Yi Wu, Wei Wu) PDF