At the 2025 IEEE Congress on Evolutionary Computation (IEEE CEC 2025) recently held in Hangzhou, China, the team from DSAI received the prestigious Best Paper Award.
The IEEE CEC is a prominent international conference in the field of evolutionary computation, providing a platform for researchers and practitioners worldwide to share research findings and exchange academic ideas. As one of the three flagship conferences of the IEEE Computational Intelligence Society (IEEE CIS), IEEE CEC began in 1994, initially held as the "IEEE Symposium on Evolutionary Computation" in Orlando, Florida, USA. Over the years, IEEE CEC has grown into a significant event in the evolutionary computation community, attracting scholars from around the globe to explore cutting-edge topics and inspire academic exchanges. Notably, this year's IEEE CEC was hosted independently in China for the first time, drawing participants from 145 countries and regions worldwide. The conference delivered a vibrant and engaging academic feast for all attendees.
The award-winning paper, titled “A Theoretical Analysis of Evolutionary Transfer Optimization”, was co-authored by DSAI postdoctoral fellows Dr XUE Xiaoming, Dr FENG Yinglan, Dr LIU Rui, and DSAI Head and Chair Professor of Computational Intelligence Prof. TAN Kay Chen, along with Prof. FENG Liang from Chongqing University and Prof. ZHANG Kai from Qingdao University of Technology. This study first links the three core subprocesses of analogical reasoning (“retrieval”, “mapping”, and “evaluation”) to the three key issues in knowledge transfer (“what to transfer”, “how to transfer”, and “when to transfer”). Subsequently, focusing on the concept of “similarity” in analogical reasoning, the research establishes theoretical foundations for knowledge transfer methods addressing the three key issues. By constructing composite functions, it systematically analyzes the greatest lower bounds of performance gains in analogy-driven transfer methods, thereby theoretically revealing both the potential and limitations of knowledge transfer. Finally, from the perspective of the “no free lunch theorem”, the study reveals the fundamental distinction between evolutionary transfer optimization (ETO) and traditional evolutionary optimization (EO). ETO introduces a novel perspective for embedding search biases, leveraging task relationship-based knowledge transfer operations that are designed to surpass traditional evolutionary operations, thereby enhancing the performance of EO.
This research offers a rigorous theoretical basis for analogy-based ETO and provides a macro-level lens to evaluate the capabilities and boundaries of knowledge transfer strategies. Looking ahead, the team will continue to focus on the theoretical exploration and algorithm design of knowledge transfer in the field of optimization. Their goal is to develop intelligent optimization systems capable of autonomously inheriting and applying experience, effectively addressing diverse optimization challenges arising from problem scale expansion, environmental changes, or shifting demands.
Official website of IEEE CEC 2025: https://www.cec2025.org/
Link to the paper: https://arxiv.org/abs/2503.21156
