[论文解读] Algorithmically probable mutations reproduce aspects of evolution such as convergence rate, genetic memory, modularity, diversity explosions, and mass extinction.

Santiago Hernández-Orozco, Héctor Zenil+1 more|arXiv (Cornell University)|Sep 1, 2017

Evolution and Genetic Dynamics被引用 4

一句话总结

本文提出基于算法概率的算法突变方法，应用于二值矩阵以模拟进化过程，相较于均匀随机突变，展现出加速收敛、遗传记忆、模块化以及多样性爆发等特性。该方法更准确地复现了生物现象，如寒武纪大爆发和大规模灭绝，表明计算在进化过程中与自然选择共同发挥关键作用。

ABSTRACT

Natural selection explains how life has evolved over millions of years from more primitive forms. The speed at which this happens, however, has sometimes defied explanations based on random (uniformly distributed) mutations. Here we investigate the application of algorithmic mutations (no recombination) to binary matrices drawn from numerical approximations to algorithmic probability in order to compare evolutionary convergence rates against the null hypothesis (uniformly distributed mutations). Results both on synthetic and a small biological examples lead to an accelerated rate of convergence when using the algorithmic probability. We also show that algorithmically evolved modularity provides an advantage that produces a genetic memory. We demonstrate that regular structures are preserved and carried on when they first occur and can lead to an accelerated production of diversity and extinction, possibly explaining naturally occurring phenomena such as diversity explosions (e.g. the Cambrian) and massive extinctions (e.g. the End Triassic) whose causes have eluded researchers and are a cause for debate. The approach introduced here appears to be a better approximation to biological evolution than models based exclusively upon random uniform mutations, and it also approaches better a formal version of open-ended evolution based on previous results. The results validate the motivations and results of Chaitin's Metabiology programme and previous suggestions that computation may be an equally important driver of evolution together, and even before, the action and result of natural selection. We also show that inducing the method on problems of optimization, such as genetic algorithms, has the potential to significantly accelerate convergence of artificial evolutionary algorithms.

研究动机与目标

• 探究基于算法概率的算法突变是否能够再现收敛速率、模块化和遗传记忆等关键进化现象。
• 将算法突变与均匀分布随机突变的零假设进行比较，评估其在进化收敛速度方面的表现。
• 探讨算法突变如何解释自然进化事件，如寒武纪大爆发和三叠纪末期灭绝。
• 评估算法突变在优化任务中增强人工进化算法的潜力。

提出的方法

• 该方法将算法突变应用于源自算法概率数值近似的二值矩阵，取代进化模拟中的均匀随机突变。
• 突变基于比特串的算法概率生成，倾向于选择更简单、更易压缩的结构，而非完全随机的结构。
• 进化过程在无重组的前提下进行，聚焦于表示基因型的二值矩阵中由突变驱动的变化。
• 通过在合成数据集和小型生物数据集上比较算法突变与均匀随机突变方案的收敛速率，衡量并对比收敛速度。
• 通过追踪规律模式在世代间的持续存在与传播，分析模块化与结构保真度。
• 通过测量新结构形态的出现速率与种群崩溃事件的频率，量化多样性与灭绝动态。

实验结果

研究问题

• RQ1 基于算法概率的突变是否相比均匀分布的突变能带来更快的进化收敛？
• RQ2 由算法驱动的进化能否再现遗传记忆，即有益结构在世代间被保留？
• RQ3 算法突变在进化系统中产生模块化与结构规律性的程度如何？
• RQ4 算法突变能否解释大规模进化现象，如多样性爆发与大规模灭绝？
• RQ5 该算法突变模型在优化任务中与传统遗传算法相比表现如何？

主要发现

• 在合成与生物实例中，算法突变相较于均匀分布突变显著加速了收敛速率。
• 在算法突变下出现的模块化结构可在世代间持续保留，表现出一种遗传记忆形式。
• 规律且可压缩的结构更可能被保留与传播，表明存在一种支持进化创新的复杂性偏好。
• 该模型复现了类似化石记录中观察到的多样性爆发与灭绝事件，如寒武纪大爆发与三叠纪末期灭绝。
• 与仅依赖随机均匀突变的模型相比，该方法对生物进化的逼近更为准确。
• 将该方法应用于人工进化算法的优化问题，有潜力显著加速收敛。