×
Machine Learning – Simulating Evolutionary Techniques to Create Images
During this project, I used machine learning and evolutionary techniques to recreate three images to the
highest possible accuracy while using just 100 polygons. This project was a large success, as I achieved
accuracy of 96% in all of these recreations. This was the highest accuracy achieved in my entire
year group.
The algorithm mimics genetic evolution by iteratively mutating a population of candidate solutions,
evaluating their fitness, and breeding the most successful candidates. It learns how to recreate input
images using just a small set of geometric shapes.
I decided to create algorithms for three different evolutionary techniques in order to find the most
I decided to create algorithms for three different evolutionary techniques in order to find the most
efficient one. The algorithms I created were for random mutation, tournament selection, and roulette
selection. I evaluated their speeds by constructing a graph using matplotlib
I used evolutionary techniques such as tournament selection, weighted roulette selection and colour
mutation, as well as changing initial parameters in order to optimise the final results. A progression of
accuracy based on these factors is shown below.
Along with my practical skills in debugging and analysing algorithms, this project was also a great
opportunity to develop my dissertation writing and reflective abilities. Most of the grading was based on an essay,
where I was judged based on how I presented my work in a clear and understandable manner, providing statistics
and insight. I received a First class grade for this.