Mathematik  |  Informatik

 

Jonathan Rigo, 2005 | Ormalingen, BL

 

This paper explores the development of a computer simulation for studying evolutionary processes. A program was created to model creature evolution in a controlled environment, allowing for the observation of emergent behaviors and strategies. The simulation removes unknowns present in real-world studies by controlling all parameters, making it easier to determine cause-effect relationships. The project demonstrates how computational models can enhance understanding of evolutionary principles with applications in various fields including antibiotic resistance research.

Introduction

How to build simple neural networks that evolve under selection pressure?

Methods

The simulation is programmed using C# programming language, chosen for its balance of efficiency and ease of use. Digital creatures were placed in a grid environment where they could interact, compete for resources, and reproduce. Each creature had unique properties and behaviors controlled by a neural network.

Results

The simulation successfully demonstrated evolutionary principles in action. Creatures developed various strategies for survival, including specialized behaviors like camouflaging, group coordination, and specific hunting strategies. The more aggressive species evolved to identify and target certain prey types, while defensive species developed techniques to avoid detection or developed spikes. The program revealed how evolution proceeds incrementally in small, gradual steps, with each adaptation building upon previous successful mutations. Different «species» were observed to emerge, which formed a complex ecosystem with distinct niches and territories. The diversity of evolutionary strategies increased over time, showing how initially simple creatures could evolve to become more complex.

Discussion

While the project was successful, several limitations were present in the current implementation. The asexual reproduction model limited diversity, and the manual process of identifying species was inefficient. The custom data storage methods created challenges when adding new properties to creatures. It was sometimes difficult to distinguish between program errors and evolutionarily advantageous traits that emerged naturally. Despite these challenges, the simulation effectively demonstrated how controlled virtual environments can provide insights into evolutionary processes that would be difficult to observe in nature.

Conclusions

The project successfully created a functional evolutionary simulation that demonstrated key principles of natural selection. For future work, several improvements are proposed: implementing sexual reproduction to increase genetic diversity, developing automated tools to generate evolutionary «family trees,» and optimizing for larger and longer simulations. These enhancements would allow for more realistic and complex evolutionary scenarios, potentially yielding deeper insights into both digital and biological evolutionary processes.

 

 

Würdigung durch den Experten

Alexandre de Spindler

Jonathan Rigo hat sich in seiner Arbeit der Frage gestellt, wie künstliche neuronale Netzwerke über einen evolutionären Algorithmus auf bestimmte Aufgaben spezialisiert werden können. Dabei hat er ein interessantes System umgesetzt, welches eine Evolution von Kreaturen simuliert und dessen Verhalten untersucht.

Er hat damit gezeigt, dass die Idee einer Evolution für solcherart Aufgaben geeignet ist, sein System kann zur Veranschaulichung der evolutionären Funktionsweise genutzt werden (z.B. im Unterricht), und ausserdem hat er eine neuartige Anwendung im Bereich der Konversations-KI ausgearbeitet und präsentiert.

Prädikat:

gut

 

 

 

Gymnasium Liestal
Lehrerin: Dr. Robyn Steiner-Curtis