Haitao Jia
+86-17320056755 | 23120356@bjtu.edu.cn | haitao1101.github.io
EDUCATION
Beijing Jiaotong University
M.Sc in Computer Science
Grade: 86.3%
Grade: 86.3%
Sep. 2023 - Present
Beijing, China
Beijing, China
Hebei University of Technology
B.Sc in Computer Science
GPA: 3.70/4.00
GPA: 3.70/4.00
Sep. 2019 - Jun. 2023
Tianjin, China
Tianjin, China
PUBLICATIONS
[C.1] Haitao Jia, et al. (2024). Mobile-Agent-v2: Mobile Device Operation Assistant with Effective Navigation via Multi-Agent Collaboration. In NeurIPS 2024.
[S.1] Haitao Jia, et al. (2023). AMBER: An LLM-free Multi-dimensional Benchmark for MLLMs Hallucination Evaluation. In submission.
RESEARCH EXPERIENCE
Dynamic Information Supplementation for Mobile AI Agents
Lenovo Research AI Lab | Advisor: Dr. Likang Wu, Dr. Ming He, Prof. Jitao Sang
- Developed a dynamic information supplementation framework for mobile multi-agent systems.
- Implemented the complete framework in Python and conducted performance testing on mobile devices.
- Fine-tuned the foundational model to optimize its adaptability to diverse mobile user tasks.
Dec. 2024 - Present
Beijing, China
Beijing, China
Mobile-Agent-v2: A Multi-Agent Architecture for Mobile Device Operation Assistance
BJTU ADAM Lab and Alibaba Group | Advisor: Dr. Junyang Wang, Prof. Jitao Sang
- Designed and implemented a multi-agent architecture comprising three specialized agents: a planning agent, a decision agent, and a reflection agent.
- Developed the full multi-agent system in Python, incorporating models and tools for screen grounding and operation execution.
Feb. 2024 - Jun. 2024
Beijing, China
Beijing, China
Fuzzy Instruction Analysis for Mobile AI Agents
BJTU ADAM Lab | Advisor: Prof. Jitao Sang
- Constructed a mobile task dataset with fuzzy instructions to enhance task comprehension.
- Fine-tuned the base language model using the dataset, enabling the model to effectively identify and address fuzzy instructions and incomplete information.
Aug. 2024 - Oct. 2024
Beijing, China
Beijing, China
Generative AI Security Competition (Attack Track)
Institute of Information Engineering, Chinese Academy of Sciences | Advisor: Dr. Yifei Gao
- Developed a transferable, context-aware large model jailbreak attack framework using prompt engineering, integrating context learning and similarity retrieval techniques to bypass security measures.
Jun. 2024 - Aug. 2024
Beijing, China
Beijing, China
An LLM-Free Multi-Dimensional Benchmark for Evaluating MLLM Hallucination
BJTU ADAM Lab | Advisor: Dr. Junyang Wang, Prof. Jitao Sang
- Contributed to the development of a benchmark to assess hallucinations in Multimodal Large Language Models (MLLMs), focusing on a holistic evaluation approach free from reliance on large language models.
Oct. 2023 - Nov. 2024
Beijing, China
Beijing, China
HONORS AND AWARDS
The First Prize National Scholarship
Beijing Jiaotong University4th in the Preliminary, 7th in the Final of the Generative AI Security Competition
Institute of Information Engineering, Chinese Academy of SciencesEmerson Scholarship
Hebei University of TechnologyThe First Prize in the Big Data Challenge for Universities in China
Hebei University of Technology
SKILLS
- Language: Chinese (native), English (IELTS Academic 7.0)
- Research Skills: Local deployment of LLMs, Supervised fine-tuning of LLMs, LLM-based Agent construction, Prompt engineering
- Programming: Python