Evolution and learning in heterogeneous environments

A real-world environment is complex and non-uniform, varying over space and time. This thesis demonstrates the impact of such environmental heterogeneity upon the ways in which organisms acquire information about the world, via a series of individual-based computational models that apply progressively more detailed forms of environmental structure to understand the causal impact of four distinct environmental factors: temporal variability; task complexity; population structure; and spatial heterogeneity. We define a baseline model, comprised of an evolving population of polygenic individuals that can follow three learning modes: innate behaviour, in which an organism acts according to its genetically-encoded traits; individual learning, in which an organism engages in trial-and-error to modify its inherited behaviours; and social learning, in which an individual mimics the behaviours of its peers. This model is used to show that environmental variability and task complexity affect the adaptive success of each learning mode, with social learning only arising as a dominant strategy in environments of median variability and complexity. Beyond a certain complexity threshold, individual learning is shown to be the sole dominant strategy. Social learning is shown to play a beneficial role following a sudden environmental change, contributing to the dissemination of novel traits in a population of poorly-adapted individuals. Introducing population structure in the form of a k-regular graph, we show that bounded and rigid neighbourhood relationships can have deleterious effects on a population, diminishing its evolutionary rate and equilibrium fitness, and, in some cases, preventing the population from crossing a fitness valley to a global optimum. A larger neighbourhood size is shown to increase the effectiveness of social learning, and results in a more rapid evolutionary convergence rate. The research subsequently focuses on spatially heterogeneous environments, proposing a new method of constructing an environment characterised by two key metrics derived from landscape ecology, “patchiness” and “gradient”. We show that spatial complexity slows the rate of genetic adaptation when movement is restricted, but can increase the rate of evolution for mobile individuals. Social learning is shown to be particularly beneficial within heterogeneous environments, particularly when mobility is restricted, suggesting that phenotypic plasticity may act as a substitute for mobility.

现实环境是复杂的，不一致的，这一论点在环境异质性上通过一系列基于个体的计算模型来表明了这种环境异质性的影响异质性。 我们定义了一个基线模型，由多基因个体的进化人群组成，可以遵循三种学习模式：生物体根据其一般编码的特征行动，其中有机体从事试验和纠正，以修改其遗传和社会学习； 该模型用于表明环境变异性和任务复杂性会影响每个学习模式的自适应成功，而社会学习只是在中位数变异性和复杂性的环境中作为一种主导策略，表明个人学习是唯一的占主导地位的策略。适应不良的个体人口的特征。 以K规范图的形式引入人口结构，我们表明，有界和僵化的邻里关系会对人口产生有害影响，从而降低其进化率和同等的适应性，在某些情况下，可以防止人口越过健身山谷，以增加较大的邻居大小，从而增加了社交学习的有效性。 随后，研究着重于空间异质性环境，提出了一种新的方法，该环境的特征是从景观生态学中得出的两个关键指标，“斑点”和“梯度”，我们表明，当运动限制时，空间的复杂性会减弱遗传适应的速度。特别是当迁移率受到限制时，表明表型可塑性可以替代迁移率。