UBDA Research Seminar: A Multi-Level-Agent Framework for Agentic AI

Abstract:
This study will present a Multi-Level-Agent framework that dynamically decomposes complex tasks into simple, specialized sub-tasks executed by intelligent nodes. Unlike traditional workflows (rigid) or general-purpose agents (long and uncontrollable chains), this framework achieves flexibility and scalability. Its key features include: (1) Standardized Node Abstraction: agents and tools are uniformly defined as “nodes,” each restricted to invoking only a few lower-level nodes. (2) Multi-Level Task Decomposition: large tasks are split into numerous lightweight sub-tasks, many solvable by small models. (3) Lightweight Communication: nodes exchange only data keys or file addresses, with results stored in a shared task workspace. (4) Reflection & Planning: base nodes integrate thinking (monitoring and re-planning) and judging (quality reflection). To evaluate the framework, we built InfiReseacher, a research assistant system. Compared to AI-Scientist V2 and Zochi, it achieved the best scores across seven criteria, including perfect marks (5/5) in Clarity and Presentation. These results show that the Multi-Level-Agent framework enables superior performance for long-horizon complex tasks. 
 
Language
English
 
Speaker
Prof. Ming LI is currently an Assistant Professor at Department of Industrial and Systems Engineering (ISE), Assistant Director of PolyU Academy for Artificial Intelligence (PAAI) and Associate Director of Artificial Intelligence Research Centre @PolyU-Daya Bay Technology and Innovation Research Institute, The Hong Kong Polytechnic University. 
 
Prior to joining ISE, Prof. LI was a Lecturer at School of Intelligent Systems Science and Engineering, Jinan University and Post-Doctoral Fellow at the HKU-ZIRI Laboratory for Physical Internet (pi-Lab), The University of Hong Kong. He received his PhD degree and master’s degree (Distinction) from Department of Industrial and Manufacturing Systems Engineering, The University of Hong Kong, in 2018 and 2013 respectively. He also obtained dual bachelor’s degrees in Computer Science (major) and Finance (minor) at South China University of Technology in 2012. He has conducted research projects in the field of Generative AI, Blockchain and Cyber-Physical System for Smart Manufacturing and Logistics with substantial government and industrial grants over 20 million HK dollars. He has published more than 100 papers on leading international journals and top conferences, including IEEE T-II, IJPE, JMS, RCIM, and etc. He has also obtained more than 40 patents and copyrights. He is the Core Member of the 2019 Guangdong Special Support Talent Program – Innovation and Entrepreneurship Leading Team. Now he is senior member of IEEE and CCF, a member of HKIE, ASME, IISE and IET, as well as the co-founder of CommaTech, a leading logistics SaaS provider, and InfiX AI, a start-up generative AI provider funded by ITC RAISe+. 

Registration & Enquiries

The seminar is free of charge with maximum of participants of 150. For registration, please complete the online enrollment form. Successful applicants will be notified before the event. For enquiries, please contact Dr Vincent Ng via vincent.ng.comp@polyu.edu.hk.