Qiwei Ma(马启伟)

Qiwei Ma(马启伟)

Shenzhen University

I’m currently a master’s student at Shenzhen University, majoring in artificial intelligence. My research interests lie in Large Language Models (LLMs), multi-agent systems, and their applications, with particular emphasis on education.

Download my CV here.

🔬 Research Experience

Multi-Agent Conversational AI for EFL Speaking Practice (Jan 2025 - Present) — Supervisor: Dr. Zhang (SZU)

  • Proposed a multi-agent system (MAS) for EFL speaking practice.
  • Completed development of seven specialized agents (preprocessing, response generation, dialogue supervision).
  • Explored mechanisms of MAS superiority, confirming synergistic effects of integrated features.
  • Found MAS outperforms SAS in oral proficiency gains (p = 0.049) and grammatical accuracy (p = 0.016) via a 4-week controlled experiment with 32 university EFL learners.
  • Observed MAS led to 26% more practice sessions, 15% longer utterances, and a 70% reduction in repeated grammatical errors.

Reasoning for Table Manipulation (Mar 2025 - Present) — Supervisors: Dr. Yang, Dr. Tan (SIAT, CAS)

  • Proposed an end-to-end LLM to manipulate tabular structures via structured reasoning.
  • Constructed a benchmark covering five core tasks: table splitting/merging, wide-to-long conversion, semi-structured field parsing, and row/column generation.
  • Completed two-stage training (SFT on reasoning traces + GRPO optimization), achieving SOTA performance among 7B-scale table-specific models.
  • Investigated structural integrity challenges and found column-level accuracy often exceeds row-level accuracy due to sensitivity to missing fields.

🔧 Others

Differential Privacy Image Generation (Mar 2024 - Apr 2025) — Supervisor: Dr. Zhang (SZU)

  • Proposed a differential privacy framework for image generation to eliminate gradient clipping bias and improve training stability.
  • Introduced reconstruction loss and noise injection during generator upsampling stages to enhance data utility and image diversity.
  • Designed multi-component training pipeline integrating generator, discriminators, classifier, and encoder with gradient sanitization mechanisms.
  • Achieved SOTA performance on MNIST and Fashion-MNIST, surpassing baseline methods in Inception Score, Frechet Inception Distance, and downstream classification accuracy.

Intelligent Annotation and Feedback System for English Writing (Apr 2025 - Present) — Supervisor: Dr. Zhang (SZU)

  • Developed LLM-IAF (LLM-based Intelligent Annotation & Feedback System), a mobile application for automated English writing evaluation targeting junior high school students.
  • Implemented dual-engine AI workflow combining semantic evaluation and visual grounding to provide immediate, visualized feedback with error localization on handwritten essays.
  • Conducted experimental study with students, comparing experimental group using LLM-IAF versus control group with traditional instruction over four weekly writing tasks.
  • Achieved significant improvement in writing performance for experimental group compared to control group, with substantial gains in learning engagement and writing self-efficacy.
  • Demonstrated strong AI-teacher score correlation and good agreement; students reported high satisfaction with system’s usefulness and ease of use.

📝 Publications

  • J. Zhang, Qiwei Ma, Y. Zhang, and X. Cao, “Multi-agent vs. single-agent AI for EFL speaking practice: A controlled experiment with hybrid input, contextual dialogue, and proficiency-adaptive feedback,” Educational Technology & Society (ET&S), 2025(Accepted)

📖 Educations

  • 2023.09 - now, Master, Shenzhen Univeristy, Shenzhen.
  • 2018.09 - 2022.06, Undergraduate, Ningxia Univeristy, Yinchuan.

💻 Internships