Evolutionary computing, including evolutionary algorithms and swarm intelligence algorithms, often experience difficulty on solving complex optimization problems that are imbued with highly time-consuming fitness functions and constrains (such as high-fidelity analysis & simulation codes) since such a paradigm often require many thousands of fitness evaluations in order to arrive at solutions that are of reasonable qualities. The use of approximation techniques is currently deemed as a practical way to augment evolutionary computation in addressing such complex problems that are highly computationally expensive.

Surrogate models are typically data-centric approximation models that mimic the behavior of the compute-intensive fitness functions and constraints as closely as possible while being computationally cheaper to evaluate. Since the modeling and design optimization cycle time is roughly proportional to the number of calls to the computationally expensive fitness function, it is now becoming a common practice for surrogate(s) to replace in part of the original solvers when engineering reliable and high quality products, due to the high commercial pressure faced in today's rapid growing competitive global economy. However, due to the curse of dimensionality, it is very difficult, if not impossible to train accurate surrogate models. Thus, appropriate evolution control or model management techniques, memetic strategies and other schemes are often indispensable. In addition, modern data analytics involving advance sampling techniques and learning techniques such as semi-supervised learning, transfer learning and active learning are highly beneficial for speeding up evolutionary search while bringing new insights into the problems of interest. New research advances, both in theory and applications on surrogate-assisted evolutionary algorithms are thus in high demand in both academia and industry. The current task force thus presents an attempt to fill this gap.

Goals:

The main goal of this task force is to promote the research on crafting novel efficient evolutionary algorithms for solving computationally expensive optimization problems. Furthermore, this task force aims at providing a forum for academic and industrial researchers to explore future directions of research and promote emerging evolutionary optimization techniques to a wider audience in the society of computational sciences

Scope:

The scope of this task force includes the following topics:

• Surrogate-assisted evolutionary optimization for computationally expensive problems
• Adaptive sampling using machine learning and statistical techniques
• Surrogate models management in evolutionary optimization
• Data-driven optimization using big data and data analytics
• Knowledge acquisition from data and reuse for evolutionary optimization
• Computationally efficient evolutionary algorithms for large scale and/or many-objective optimization problems
• Real world applications including multidisciplinary optimization

Planned Activities:

• Propose a Special Session on “Data-Driven Evolutionary Optimization for Computationally Expensive Problems” for SSCI 2016
• Propose a Special Session on “Efficient Non-dominated Sorting and Pareto Approaches to Many-Objective Optimization” at WCCI 2016
• Propose a Special Session on “Computationally Expensive Optimization” at WCCI 2016