Evolution inspired development of Multi-level Selection Genetic Algorithm for general applications

遗传算法被用于一系列不同的应用中，包括：航天器天线设计，电脑游戏中的人工智能，确定哪些基因导致疾病，以及金融投资的投资组合选择。有一个充满活力的研究人员和工业家社区，他们正在开发新的方法来提高这些算法的性能，从而允许解决更复杂的问题，并允许它们在新的应用中使用。一旦这种方法在南安普顿大学被开发出来(10.1088/17483190/aad2e8和10.1016/j.Swevo.2018.09.005)，它就表现出作为一般解算器的领先性能，这意味着它在广泛的问题上表现得很好，并显示出特别好的种群多样性保护。它实现了多层选择理论和协同进化方法来提高算法的性能，并在许多基准问题和现实应用中表现出了领先的性能。然而，这种方法仍然是新的，有许多可以改进的地方，所以这个项目将专注于它的继续发展。这种算法的灵感已经在很大程度上建立在进化论的基础上，这种理论基于这样的想法，即加速自然界进化速度的机制，也同样会加快遗传算法的速度。在这个项目中，我们将从表观遗传学、跨代可塑性、社会进化和多级选择理论中获得灵感，以测试遗传算法的性能是否可以得到类似的提高。将当代进化理论与已经非常高效的遗传算法相融合，有可能给进化计算带来革命性的变化。

Genetic algorithms are used in a range of diverse applications including: spacecraft antenna design, AI in computer games, determining which genes cause illness and portfolio selection for financial investments. There is a vibrant community of researcher and industrialists developing novel methods to improve the performance of these algorithms allowing more complex problems to solve and allowing their use in new applications. Once such method was developed in the University of Southampton (10.1088/1748-3190/aad2e8 and 10.1016/j.swevo.2018.09.005) which shows leading performance as a general solver, meaning it performs well across a wide range of problems, and shows particularly good preservation of the diversity of a population. It implements the Multi-Level Selection Theory and co-evolutionary approaches to improve the performance of the algorithm and already shows leading performance on a number of benchmarking problems and real world applications. However, the method is still new and there are a number of improvements that can be made and so this project will focus on its continued development. The inspiration for this algorithm is already heavily based in evolutionary theory based on the idea that mechanisms that accelerate the rate of evolution in the natural world, will similarly accelerate the speed of the genetic algorithm. In this project we will take inspiration from epigenetics, transgenerational plasticity, social evolution and multi-level selection theory to test if the performance of the genetic algorithm can be similarly enhanced. Fusing contemporary evolutionary theory with an already highly efficient genetic algorithm has the potential to revolutionise evolutionary computation.