Dongda Li
PhD Researcher in Vision-Language Models and Multimodal Learning
I am a PhD student in Computer & Information Science & Engineering at Syracuse University, studying vision-language models, compositional reasoning, multimodal robustness, and reliable evaluation. My earlier work spans reinforcement learning, robotics, wireless sensor networks, and autonomous systems.
Research Interests
- Vision-Language Models
- Multimodal Robustness
- Compositional Reasoning
- Embodied AI Evaluation
- Reinforcement Learning
- Autonomous Systems
Education
Syracuse University
PhD in Computer & Information Science & Engineering
Advisor: Senem Velipasalar
Xi’an University of Technology
B.E. in Automation
Current Research
PhD Researcher
Syracuse University, Department of Electrical Engineering and Computer Science
Affiliated with the Institute for Artificial Intelligence
Vision-Language Models and Multimodal Reliability
I study how multimodal models represent and use compositional information, particularly object attributes, relations, and role binding. My work analyzes CLIP, SigLIP 2, and related frozen dual encoders to identify where role information remains accessible in regional features and where it becomes inaccessible to deployed similarity scoring. I develop controlled probing and evaluation pipelines across SugarCrepe, SugarCrepe++, Winoground, synthetic spatial pairs, and real-image role-binding data, along with efficient crop-based and self-attention readouts.
Embodied AI Evaluation
I also study the reliability of automatic robot-policy evaluators, including reward models, VLM judges, and multimodal scoring systems. This work evaluates automatic methods on 5,106 real-robot episodes against 3,284 double-blind human comparisons, with an emphasis on human-ranking agreement, uncertainty, implementation sensitivity, and reproducibility.
Publications
- Dongda Li, Zhaoquan Gu, Yuexuan Wang, Changwei Ren, and Francis C.M. Lau. “One Model Packs Thousands of Items with Recurrent Conditional Query Learning.” Knowledge-Based Systems, 235:107683, 2022. Reinforcement-learning-based neural combinatorial optimization.
- Zhaoquan Gu, Dongda Li, Nadra Guizani, Xiaojiang Du, and Zhihong Tian. “An Aerial-Computing-Assisted Architecture for Large-Scale Sensor Networks.” IEEE Wireless Communications, 28(5):43-49, 2021.
- Dongda Li, Yuexuan Wang, Zhaoquan Gu, Tong Shen, Tianhao Wei, Yongqin Fu, Heming Cui, Mingli Song, and Francis C.M. Lau. “Adler: A Resilient, High-Performance and Energy-Efficient UAV-Enabled Sensor System.” HKU Technical Report TR-2018-01.
- Tong Shen, Yuexuan Wang, Zhaoquan Gu, Dongda Li, Zhen Cao, Heming Cui, and Francis C.M. Lau. “Alano: An Efficient Neighbor Discovery Algorithm in an Energy-Restricted Large-Scale Network.” IEEE International Conference on Mobile Ad-hoc and Sensor Systems, 2018.
Research Funding and Computing Allocations
NSF ACCESS-CI Discover Allocation - Principal Investigator
Efficient Evidence Bottlenecks for Compositional Vision-Language Understanding
Awarded an ACCESS-CI Discover research computing allocation supporting experiments on efficient compositional vision-language models.
Product Engineering & Open Source
Mindwtr - Founder & Principal Developer
I founded Mindwtr and own its product and engineering direction end to end. It is an open-source, local-first, cross-platform productivity application with 30,000+ users, ~2,000 daily active users, and 2,000+ GitHub stars. I shape the roadmap from real user feedback, design and build features across desktop, mobile, and web, and manage releases and the open-source community. The engineering spans synchronization, optional AI functionality, MCP/REST integrations for agent and automation workflows, and multi-platform deployment.
Experience
Goertek Inc.
Robotics Engineer · Algorithm Group, R&D
Robotic perception and control, SLAM, motion planning, navigation, and trajectory planning.
Syracuse University
Teaching Assistant
CPS 181 Introduction to Computing; CIS 454 Software Implementation; CIS 675 Design and Analysis of Algorithms.
Guangzhou University
Research Assistant · Cyberspace Institute of Advanced Technology
Reinforcement learning, neural combinatorial optimization, meta-learning, and learning-based combinatorial decision making.
The University of Hong Kong
Research Assistant · Department of Computer Science
Robotic networks, wireless sensor networks, and UAV-enabled sensing.
Beijing Institute of Technology
Research Assistant · UAV Autonomous Control Institute
UAV control systems, navigation, and data fusion.
Beijing Zhonghangzhi Inc.
Robotics Engineer · Control Group, R&D
Robotic navigation and control.
Selected Projects
- Collaborative analysis and autonomous decision technology for intelligent manufacturing big data - National Key R&D Program of China (2018-2021).
- Software-defined wireless sensor network system (2017-2018).
- UAV-enabled sensor system (2017).
- MBZIRC 2017 International Robotics Challenge; core member of the champion team.
- Autonomous takeoff and landing of an intelligent quadrotor - undergraduate thesis (2014-2015).
Technical Skills
- Programming: Python, C/C++, TypeScript, Rust
- Machine Learning: PyTorch, Transformers, reinforcement learning, multimodal evaluation
- Agentic AI Systems: LLM tool use, agent workflows, MCP, REST APIs
- Systems: Linux, CUDA-based ML workflows, Git, HPC and cluster-based experimentation
- Robotics: ROS, perception, SLAM, navigation, motion planning, and control
Service and Awards
- Reviewer, Journal of Contemporary Mathematics, 2024
- Reviewer, IEEE Journal on Selected Areas in Communications, 2022
- Reviewer, IEEE International Conference on Intelligent Robots and Systems, 2020
- Core member of the MBZIRC 2017 champion team
- Grand Prize, Eighth Shaanxi Province Outstanding Graduation Design, 2015
- Volunteer translator (English to Chinese), CS 285 Deep Reinforcement Learning
Languages
- Chinese: native
- English: IELTS 7.0