Super-Individuals - to Balance Speed and Accuracy of Individual-based Models on Demand

Wilsdorf, Pia (2018) Super-Individuals - to Balance Speed and Accuracy of Individual-based Models on Demand. Masters thesis, Institute of Computer Science, University of Rostock.

Full text not available from this repository.

Abstract

Individual-based modelling and simulation has been established as a key tool in ecology, as it allows for very fine-grained analyses of dynamic systems. For example, in fisheries building individual-based models aims at understanding the various factors that influence stock sizes of fish populations. However, despite having many advantages over other modelling paradigms such as populations-based approaches, their major drawback is their poor scalability. Consequently, alternative modelling and simulation techniques are required, such as the so-called super-individual approaches. These techniques approximate subpopulations of homogeneous entities by a representative individual. Such a model abstraction allows for shorter execution times, however, the reduction of computation time is bought by an increasing inaccuracy of the simulation results. Thus, the main challenge lies in balancing speed and accuracy. While existing super-individual approaches are mostly static, in this thesis we aim at dynamic super-individuals, i.e., allowing them to reorganise during the simulation to prevent excessive homogenisation within the subpopulations and to compensate the induced approximation error. We develop two approaches that differ in the way they handle the dynamic aspects, and implement them exemplarily in the ML-Rules modelling and simulation tool. An evaluation and direct comparison of both approaches in terms of speed and accuracy demonstrates how the compromise between speed and accuracy can be steered by the modeller through parameters based on expert knowledge. The results indicate that dynamic super-individuals are a first decisive step towards more efficient execution of fish population models. However, the evaluation also reveals numerous long-term goals that need solving before the approach may be applied in practice

Item Type: Thesis (Masters)
Projects: ESCeMMo