Header Image

IEEE CIS TASK FORCE ON

LARGE MODELS FOR INDUSTRIAL OPTIMIZATION

MOTIVATION

Industrial systems are becoming increasingly complex due to the integration of advanced technologies, the expansion of interconnected networks, and the rising demand for higher efficiency and productivity. This growing complexity necessitates more sophisticated modeling and management strategies to ensure reliability and adaptability in ever-changing environments. Industrial optimization refers to the application of various optimization techniques and methods to improve the efficiency of processes and operations within industrial systems. The goal is to enhance system performance, making it more intelligent, efficient, cost-effective, and productive. The fourth industrial revolution involves an unprecedented integration of sensing, computing, and control, with industrial optimization playing a critical role in enhancing system intelligence.

Significant efforts have been made in industrial optimization within the context of computational intelligence. For example, optimized scheduling and planning by evolutionary algorithms can significantly reduce the resource costs of the energy system in ethylene plants; using a reinforcement learning-aided decision procedure can automatically select optimal operational indices for unit processes in an industrial plant; learning and optimization from data can shorten the period of pharmaceutical research. Although traditional computational intelligence methods, such as evolutionary algorithms, have addressed numerous optimization problems in the pursuit of industrial intelligence, they often encounter several pervasive shortcomings, including limited generalizability, an inability to fully exploit the intrinsic properties of the problems, and a dependence on prior domain knowledge.

Recently, large models (LMs) have attracted extensive attention and demonstrated great success in a wide range of real-world applications, including computer vision, natural language processing, and speech recognition. LMs refer to advanced machine learning frameworks that consist of a significant number of parameters and complex architectures, enabling them to process and learn from vast datasets. While large language models are a prominent example, particularly in natural language processing tasks, LMs also include large vision models, generative models, reinforcement learning models, and multimodal models. These models leverage their capacity to learn intricate patterns from large datasets, enabling them to effectively capture the underlying complexities of industrial systems. The advantages of employing LMs in industrial optimization include enhanced predictive accuracy, improved decision-making capabilities, and the ability to generalize across diverse scenarios, ultimately leading to more efficient resource allocation, reduced operational costs, and the potential for innovative solutions in dynamic environments. By harnessing these capabilities, industries can achieve significant improvements in productivity and adaptability.

GOALS

While the process of achieving industrial intelligence involves addressing numerous optimization problems, including scheduling, resource allocation, supply chain management, and predictive maintenance, this diversity has resulted in researchers often working in isolation, hindering effective resource integration. One of the primary goals of our task force is to systematically summarize and categorize these industrial optimization issues over the next two years, ultimately developing a comprehensive set of standard problems for public dissemination.

Furthermore, although research on LMs is rapidly advancing, investigations into their use for enhancing evolutionary algorithms are still in their infancy, particularly regarding applications to industrial challenges. Our task force aims to bridge this gap by fostering collaboration among scholars and practitioners in the field. We will explore the integration of LMs with evolutionary algorithms to create hybrid approaches that leverage the strengths of both. To validate our approaches, we will also collaborate with industry partners to implement and test these hybrid algorithms in real-world scenarios.

Through these efforts, we aim to create a robust research community focused on leveraging large models to enhance evolutionary algorithms, addressing complex industrial challenges and driving innovation in optimization practices. By aligning our research with actual industrial needs, we seek to ensure that our work has a meaningful impact on both the academic and industrial landscapes.

SCOPS

The scope of this task force includes, but is not limited to:

  • LM Integration with Simulation Engines: Investigating methods for embedding LMs and other large AI models into traditional simulation engines, enhancing their ability to predict complex system behaviors. For example, integrating LMs with computational fluid dynamics (CFD) simulations to better predict fluid behaviors under varying conditions.
  • LM-Enhanced Evolutionary Algorithms: Developing new algorithms that utilize the computational strengths of LMs for optimization tasks. For instance, designing hybrid optimization techniques that combine LM-driven insights with evolutionary algorithms for optimizing manufacturing processes.
  • Data-Driven Evolutionary Optimization: Addressing the issue of time-consuming simulations through collected historical data.
  • Predictive Maintenance Scheduling: Applying LMs to predict equipment failures and optimize maintenance schedules. This could involve using LMs to analyze historical maintenance data and operational parameters to schedule proactive maintenance for industrial machinery.
  • Real-World Application Case Studies: Conducting detailed case studies to validate the effectiveness of large model-enhanced simulations. This could involve collaborating with industries to implement and assess the impact of these advanced simulations in real-world scenarios.

Activities

Chair

  • Shulei Liu, Xidian University, China
    Shulei Liu received the M.Eng. degree from Shandong Normal University, Jinan, China, in 2018, and the Ph.D. degree from Xidian University, Xi’an, China. He is currently a postdoctoral researcher at Xidian University. His research interests include large-scale multi-objective optimization, data-driven combinatorial optimization, large model-based evolutionary optimization, and real-world industry problems.

Vice Chair

  • Qingfu Zhang, City University of Hong Kong, China
    Qingfu Zhang received the B.S. degree in mathematics from Shanxi University, Taiyuan, China, in 1984, and the M.S. degree in applied mathematics and the Ph.D. degree in information engineering from Xidian University, Xi’an, China, in 1991 and 1994, respectively. He is a Chair Professor of Computational Intelligence at the Department of Computer Science, City University of Hong Kong. His main research interests include evolutionary computation, multiobjective optimization, machine learning and their applications. Dr. Zhang is an Associate Editor of the IEEE Transactions on Evolutionary Computation and the IEEE Transactions on Cybernetics. He has been a Web of Science highly cited researcher in Computer Science for six times.

Members

  • Handing Wang, Xidian University, China, hdwang@xidian.edu.cn
  • Sanaz Mostaghim, Otto-von-Guericke University Magdeburg, Germany, sanaz.mostaghim@ovgu.de
  • Fei Liu, City University of Hong Kong, China, fliu36-c@my.cityu.edu.hk
  • Chao Qian, Nanjing University, China, qianc@nju.edu.cn
  • Tonghao Wang, Xidian University, China, wangtonghao@xidian.edu.cn
  • Fernando Jiménez, University of Murcia, Spain, raquel.espinosa@um.es
  • Terence L. van Zyl, University of Johannesburg, South Africa, tvanzyl@uj.ac.za
  • Rudolf Kruse, Otto-von-Guericke University Magdeburg, Germany, rudolf.kruse@ovgu.de
  • Takumi Sonoda, Yokohama National University, Japan, sonoda-takumi-gp@ynu.jp
  • Christoph Steup, Otto von Guericke University, Germany, christoph.steup@ovgu.de
  • Junhao Huang, Victoria University of Wellington, New Zealand, junhao.huang@ecs.vuw.ac.nz
  • Jialong Li, Waseda University, Japan, lijialong@fuji.waseda.jp
  • Hao Hao, Shanghai Jiao Tong University, China, haohao@sjtu.edu.cn
  • Zhenshou Song, Victoria University of Wellington, New Zealand, Songzhenshou@gmail.com