Emergence and Open-ended Evolution in Genetic Programming

遗传编程中的涌现和无限制进化*涌现在科学和技术中的作用才刚刚开始变得非常明显。长期以来，新出现的现象一直局限于特定领域的专家讨论，但随着科学和工程领域引入大规模模拟方法，这种情况即将改变，因为它将失去其“奇异特征”，变得清晰可见：这是一种在众多学科中具有解释力的普遍现象。另一方面，新颖性和创新长期以来一直被认为是科学技术进步的驱动力。然而，它们的建模和形式化一直很困难，因为时间在任何新颖性概念中发挥着重要的作用，而且在正式系统中捕捉这一概念存在技术困难。*我们将使用生物启发计算的前沿方法之一--遗传编程来研究计算系统中的新奇创造和涌现现象。特别是，我们感兴趣的是如何在这样的系统中实现持续的创造力过程，不断地产生新颖性。我们还将尝试将系统中新实体的出现正规化，将遗传编程作为设置适当条件的实验室。*通过从达尔文进化过程中获得灵感，遗传编程不仅允许解决单个问题实例，就像在组合优化中一样，还允许使用导致适应性有机体和更高级别有机体的智能的自然“配方”来生成解决问题的实体。随着摩尔定律的持续有效，未来十年将展示生物启发算法产生的与人类竞争的结果，特别是从进化中收集到的结果。*遗传编程是人工智能发展的一部分，它是由同样的力量推动的--(1)所有科学和技术领域的大量数据的可用性；(2)计算更复杂算法的处理周期的持续增长；(3)神经网络的深度学习等新范式，它们彻底改变了整个应用领域。*我们将把遗传编程作为我们的“实验室”，探索一个算法系统中的新颖性、创新性、涌现性和创造力，并将其应用于多个领域：机器学习和人工智能、大数据分析、计算科学和数学。除了对遗传编程算法的方法论改进外，我们还将研究它们在两个重要应用领域的行为，即深度学习神经网络的构建和用于修复计算机代码中的错误的代码补丁的演变。

Title Emergence and Open-ended Evolution in Genetic Programming******The role of emergence in science and technology is just beginning to become highly visible. Emergent phenomena have long been constrained to the discussion of specialists in particular fields but with the introduction of large-scale simulation approaches to science and engineering, this is about to change in in that it will loose its "exotic character" and become visible in what it is: A widespread phenomenon with explanatory power in a vast array of disciplines. Novelty and innovation, on the other hand, have long been recognized as drivers of progress in science and technology. However, their modelling and formalization has been difficult, due to the essential rule time plays in any notion of novelty and the technical difficulties of capturing that concept in formal systems. ******We shall use one of the methods at the cutting edge of Bio-inspired Computing, Genetic Programming, to study the phenomena of novelty creation and emergence in computational systems. In particular, we are interested in how to achieve an ongoing process of creativity, a continued generation of novelty, in such systems. We further shall attempt to formalize the emergence of new entities in systems, using Genetic Programming as a laboratory in which to set appropriate conditions. ******By taking inspiration from the Darwinian process of evolution, Genetic Programming allows not only to solve individual problem instances, like in combinatorial optimization, but to use the “recipe” of Nature which led to adaptive organisms and the intelligence of higher-level organisms to generate problem-solving entities. With the continued validity of Moore's law, the next decade will demonstrate an explosion of human-competitive results from bio-inspired algorithms, and in particular from those gleaned from evolution. ******Genetic Programming is part of the upswing in Artificial Intelligence, which is driven by the same forces – (1) availability of large amounts of data in all areas of science and technology; (2) continued growth in the availability of processing cycles for calculation of ever more complex algorithms; (3) new paradigms like deep learning for neural networks that have revolutionized entire application areas.******We shall take Genetic Programming as our “laboratory” to explore novelty, innovation, emergence and creativity in an algorithmic system with applications to multiple domains: machine learning and artificial intelligence, big data analysis, computational science and mathematics. Besides methodological improvements to genetic programming algorithms, we shall examine their behaviour in two important application areas, the structuring of deep learning neural networks and the evolution of code patches for repairing bugs in computer code.