Jiayi Ma马佳艺

Research科研成果

More than ten presentations at international and national conferences on physical education, sports training and performance analysis — with one question underneath all of them: how do you bring computer vision and language models onto the fencing piste without losing what a coach knows?累计进行国际及全国会议报告 10 余次,主题涵盖体育教育、运动训练与运动表现分析。所有工作指向同一个问题:如何把计算机视觉与大语言模型带上击剑剑道,而不丢失教练真正掌握的东西?

01

In progress在研课题

M.S. thesis · Beijing Sport University · 2024 –硕士学位论文 · 北京体育大学 · 2024 –

Injury Risk Assessment in Youth Sabre Fencing Based on Computer Vision and Large Language Models基于计算机视觉与大语言模型的青少年佩剑损伤风险评估研究

Approach研究思路

A five-stage pipeline — skeleton extraction (YOLOv8-Pose) → a 19-indicator biomechanics system → knowledge injection → preference alignment → automated risk assessment — producing interpretable injury-risk reports for sabre fencers aged 12–18.提出五阶段技术管线:骨架提取(YOLOv8-Pose)→ 19 项指标生物力学体系 → 知识注入 → 偏好对齐 → 自动化风险评估,为 12–18 岁青少年佩剑运动员生成可解释的损伤风险报告。

Novelty创新点

The first application of two-stage SFT + DPO alignment (Qwen2.5-7B + LoRA, LlamaFactory) to sports injury-risk assessment, validated against national-level coach evaluation.首次将 SFT + DPO 两阶段对齐方法(Qwen2.5-7B + LoRA,LlamaFactory)应用于运动损伤风险评估,并以国家级教练员评估作为验证基准。

  • YOLOv8-Pose
  • Qwen2.5-7B
  • LoRA / LlamaFactory
  • SFT + DPO
  • Markerless Motion Capture

02

Conference papers & talks会议论文与报告

Who Coaches the Coach? — Augmenting Pedagogical Decision-Making Through Expert-Aligned AI in Adolescent Sabre Training谁来教练教练?—— 以专家对齐的人工智能增强青少年佩剑训练中的教学决策

Developmental Sequence of Sabre Lunge Technique佩剑弓步技术的发展序列研究

Markerless Motion-Capture–Based Early-Warning Model for Lower-Limb Injuries in Fencing Athletes基于无标记动作捕捉的佩剑运动员下肢损伤预警研究

Deep-Learning-Based Precision Recognition and Analysis System for Sabre Techniques基于深度学习的佩剑动作精准识别与分析系统研究

Performance-Training Games Improve Fencing Techniques by Affecting Physical and Cognitive Function in High-School Athletes表现性训练游戏通过影响身体与认知功能提升高中生击剑技术

Deep-Learning-Based Accurate Recognition and Analysis of Sabre Movement基于深度学习的佩剑动作精准识别与分析

Effects of Experimental Game-Based Training on Promoting Physical Fitness of High-School Fencing Athletes实验性游戏训练法对促进高中生击剑运动员体质健康的影响

03

Applied projects应用项目

Hackathon · full stack · Aug 2026黑客松 · 全栈 · 2026.08

毫厘 · PROOF OF CALL毫厘 · PROOF OF CALL

An evidence-chain system for right-of-way calls in sabre. It does not replace the referee — it makes the call auditable. Selected for the SheNicest 1000-person hackathon, software track.佩剑对攻判罚的证据链系统。它不代替裁判判罚,只让每一次判决有据可查。入选千人 SheNicest 黑客松软件赛道。

github.com/majiayi0526/haoli-proof-of-call ↗

Software copyright (China)软件著作权

Fencing Movement Recognition Mini-Program V1.0击剑动作识别小程序 V1.0

Model training and mini-program development.动作识别模型训练与小程序开发。

Beijing Information Sci-Tech University北京信息科技大学

AI Motion Recognition for Pickleball Skill Analysis基于 AI 动作识别的匹克球运动技能分析

Computer-vision human-motion recognition applied to pickleball stroke and footwork analysis. Website development and model training.将基于计算机视觉的人体动作识别应用于匹克球击球与移动技术分析。负责网站开发与模型训练。

pits-v1.vercel.app ↗

Beijing Sport University Fencing Team北京体育大学击剑队

Expert Chain-of-Thought Simulation in Épée Training专家思维链仿真在重剑训练中的应用

Amateur épée athletes face delayed movement feedback and injury risk that is hard to spot in advance. This project reproduces a senior coach's diagnostic pathway as an executable chain of thought.业余重剑选手常面临动作反馈滞后、损伤风险难以前置识别的问题。本项目以可执行的专家思维链仿真复现资深教练的诊断路径。

Research group member · First Prize课题组成员 · 一等奖

Sports Games in Junior-High PE Teaching体育游戏在初中体育教学中的开展现状

PE Teaching & Research Award, First Prize.体育教研一等奖。

04

Technical & research skills技术与研究技能

Engineering工程与建模

  • Python
  • R
  • SQL
  • PyTorch
  • OpenCV
  • YOLOv8-Pose
  • LoRA / LlamaFactory
  • Streamlit

Method研究方法

  • SPSS
  • Inferential statistics (t-test)
  • Weighted Kappa / ICC
  • Expert-panel interview design
  • Markerless motion capture