Student at Institute of Computing Technology, Chinese Academy of Sciences, China
Chenhao Li is a dedicated Ph.D. candidate at the Institute of Computing Technology, Chinese Academy of Sciences, with a strong focus on designing lightweight and adversarially robust neural networks. His work lies at the intersection of model pruning, quantization, and test-time adaptation, aiming to accelerate deep learning models while maintaining or improving robustness. He has independently led multiple impactful research projects and published in top-tier venues such as AAAI. His contributions span both theoretical innovations and practical implementations in real-world systems, including drones, unmanned platforms, and embedded devices. Demonstrating strong academic independence and technical depth, he has received notable awards such as the Director’s Excellence Award and top academic scholarships. Chenhao’s solutions have resulted in significant efficiency gains with minimal accuracy loss, showcasing his ability to translate complex algorithms into deployable AI models. His ongoing efforts continue to push the boundaries of robust and efficient AI in edge computing and real-time applications.
Professional Profile
Education🎓
Chenhao Li is currently pursuing his Ph.D. in Computer Software and Theory at the Institute of Computing Technology, Chinese Academy of Sciences (2019–present), where he is engaged in advanced research on lightweight and robust neural network models. His doctoral studies involve extensive work in pruning, quantization, and adversarial training for deep learning. Prior to this, he completed his Bachelor of Science in Computer Science and Technology at the University of Chinese Academy of Sciences (2015–2019). Throughout his academic journey, Chenha
o has developed a solid foundation in artificial intelligence, computer vision, and model optimization techniques. His education has provided him with both the theoretical knowledge and practical skills required to innovate in areas like model compression, test-time adaptation, and real-world deployment of deep learning systems. He has continuously demonstrated academic excellence and has been recognized with scholarships and awards throughout his studies, reflecting both his dedication and strong grasp of core computing principles.
Professional Experience📝
Chenhao Li has contributed to several high-impact research projects involving model compression, object detection, and adversarial robustness. Between 2020 and 2024, he played key roles in five major projects, including drone-based object detection acceleration, lightweight model design for visible and infrared data integration, and robust model deployment for unmanned platforms. In these roles, he implemented advanced pruning and quantization techniques, reducing model parameters by up to 95% and improving inference speeds significantly—often with negligible or no accuracy loss. He also worked on deploying models on Nvidia inference chips, demonstrating strong backend and embedded systems skills. His practical experience is distinguished by his ability to apply cutting-edge algorithms to real-world challenges, particularly in embedded and edge AI contexts. Chenhao’s contributions span the full lifecycle from model design and training to deployment and optimization, emphasizing his well-rounded expertise in both theoretical research and applied AI development across diverse industrial applications.
Research Interest🔎
Chenhao Li’s research interests center on developing lightweight and robust deep learning models, especially for edge computing and safety-critical applications. He is particularly focused on neural network pruning, quantization, and adversarial robustness. His work addresses the need to balance efficiency and security in AI models by designing novel frameworks that maintain performance even under severe compression or adversarial attacks. Additionally, he explores test-time adaptation strategies that help models remain stable and accurate under domain shifts and corrupted inputs. Chenhao’s passion lies in the practical deployment of AI systems, driving his interest in designing models that can run effectivel
y on resource-constrained devices such as drones or embedded chips. His innovations, including weight reparameterization, two-stage reconstruction, and angular distance-based adaptation, reflect a deep commitment to pushing the boundaries of robustness and efficiency in modern neural networks. He also keeps a close eye on developments in large model acceleration and hardware-aware AI.
Award and Honor🏆
Chenhao Li has received multiple recognitions for his academic excellence and research achievements throughout his career. Notably, he was awarded the Director’s Excellence Award by the Institute of Computing Technology, Chinese Academy of Sciences, for his outstanding performance in research and innovation during the academic year 2022–2023. He has also earned first and second prize scholarships consistently between 2019 and 2024, recognizing his superior academic standing and contributions to the scientific community. These honors reflect both his intellectual rigor and dedication to advancing the field of computer science. His award-winning work includes significant improvements in model pruning and adversarial robustness, achieving cutting-edge results with practical deployment value. These recognitions further validate his role as a promising young researcher, with a strong trajectory toward becoming a leader in AI optimization, embedded AI systems, and robust machine learning. The accolades reinforce his commitment to high-impact, technically advanced, and socially relevant research.
Research Skill🔬
Chenhao Li possesses a robust set of research skills that span theoretical design and practical implementation of deep learning models. His core competencies include model pruning, quantization, adversarial training, test-time adaptation, and neural network optimization. He is proficient in designing reparameterization strategies for robustness, implementing efficient training pipelines, and conducting structured and unstructured pruning with minimal accuracy degradation. Chenhao has hands-on experience with deployment frameworks such as MMDetection, and has successfully deployed models on embedded and Nvidia inference chips. He excels in managing end-to-end pipelines, including data preprocessing, model compression, training acceleration, and real-time deployment. His experiments consistently achieve high accuracy and performance benchmarks while minimizing computational demands. Additionally, he is skilled in programming languages and tools such as Python, PyTorch, and CUDA. His ability to bridge academic innovation and industrial application makes him highly capable in addressing the efficiency and reliability challenges of modern AI systems.
Conclusion💡
Chenhao Li is a highly promising and technically adept researcher with strong achievements in adversarial robustness, model pruning, and real-world AI acceleration. His independent research ability, innovative methodologies, and practical implementations make him a highly suitable candidate for the Best Researcher Award. With expanded publication visibility and continued cross-domain collaboration, he is poised to become a leading expert in efficient and robust deep learning systems.
Publications Top Noted✍
-
Title: Learning Adversarially Robust Sparse Networks via Weight Reparameterization
-
Authors: Chenhao Li, Qilin Qiu, Zhe Zhang, Jing Guo, Xianglong Cheng
-
Year: 2023
-
Citations: 7