Hello, I am
你好，我是
I am an undergraduate in Software Engineering at Beijing University of Technology and a former research assistant at the Wangxuan Institute of Computer Technology, Peking University. I work on reliable evaluation and improvement of multimodal large language models.
我是北京工业大学软件工程专业本科生，曾在北京大学王选计算机研究所担任科研助理。 我的研究关注多模态大语言模型的可靠评估与生成结果优化。
My current interests include multimodal LLM evaluation, Figure-to-Text generation, agent systems, and parameter-efficient fine-tuning.
目前主要兴趣包括多模态大模型评估、图表转文本生成、 智能体系统与参数高效微调。
Research journey
科研历程
The items below describe submission experience rather than published papers. Status labels are kept explicit.
以下内容为论文投稿经历，并非已发表成果；页面对投稿状态作明确标注。
Human-aligned multimodal Figure-to-Text evaluation and refinement
面向人类对齐的多模态图表文本评估与优化
First-author manuscript currently submitted to EMNLP. It presents a generation–evaluation–refinement framework, a five-dimensional Chain-of-Evaluation protocol, and a human-verified data-construction process.
第一作者论文，目前投稿 EMNLP。工作围绕“生成—评估—优化”框架、五维度 Chain-of-Evaluation 评估机制，以及包含人工验证的数据构造流程展开。
Peer-review experience and subsequent revision
同行评审经历与后续修改
An earlier version was submitted but not accepted. The review process exposed limitations in novelty positioning, baseline-comparison fairness, the synthetic-data protocol, ablation depth, and efficiency analysis.
早期版本曾投稿但未获录用。评审意见指出了创新点定位、基线比较公平性、合成数据流程说明、 消融实验深度与效率分析等方面的问题。
Background
教育与科研
Wangxuan Institute of Computer Technology, Peking University
Conducted API-versus-local-model experiments, LoRA fine-tuning, inference benchmarking, model-loading optimization, and automated dataset-pipeline design for chart evaluation.
参与大模型 API 与本地模型对比实验、LoRA 微调、推理测试、模型加载调优，以及图表评估数据流水线设计。
Beijing University of Technology (211 University)
Core coursework: data structures, computer organization, operating systems, databases, computer networks, and C/C++ programming.
主修课程包括数据结构、计算机组成原理、操作系统、数据库、计算机网络与 C/C++ 程序设计等。
Recognition
获奖与成果
No. 202511282682.4
Mathematical Contest in Modeling
Programming competition
程序设计竞赛
China Youth Innovation and Entrepreneurship Competition
中国青年创新创业大赛
Research focus
研究方向
Post-training, instruction tuning, inference behavior, and reliable use of open-source language models.
关注大模型后训练、指令微调、推理行为，以及开源大模型的可靠应用。
Vision-language models, Figure-to-Text generation, tool-using agents, and iterative generation–evaluation–refinement.
关注视觉语言模型、图表转文本、工具调用智能体，以及生成—评估—优化的迭代机制。
Human-aligned benchmarks, fine-grained multidimensional scoring, dataset construction, and interpretable evaluation.
关注人类对齐评测基准、细粒度多维评分、数据集构造与可解释评估。
Parameter-efficient fine-tuning, LoRA, local model deployment, inference benchmarking, and loading optimization.
关注参数高效微调、LoRA、本地模型部署、推理性能测试与模型加载优化。