CEPDT
In sixth form I designed and developed a neuroevolutionary algorithm and ecosystem simulation in order to explore how adaptive behavior could emerge from biologically inspired evolutionary processes. The algorithm, Chimeric Evolution of Partially Divergent Technologies (CEPDT), was designed to model modular evolution within artificial neural architectures.
Sep 2022 - May 2023 • 8 months
Tech Stack
PythonSQLOpenCVTauriSvelte
Key Features:
- The algorithm: implemented in python, using a novel binary encoding scheme representing each neural network as a ‘genome’, where the neural network was composed of a core architecture (static topology) and supplementary architecture (dynamic topology). Custom genetic operators inspired by mutation and recombination generated variation in weights, biases, topologies, and activation functions.
- Evolutionary simulation: built a 2D agent based simulation where agents competed for resources across varying ecosystem setups, balancing energy expenditure to avoid hazards and locate resources whilst competing with others.
- Ecosystem branching: implemented a checkpoint system to pause and re-run simulations from any point with the current agent population to explore divergent evolutionary trajectories.
- Data management: designed a multi-table relational SQL database to track genomes, simulation states and recorded images, enabling branching experiments and video reconstruction.
- Image processing: integrated OpenCV to perform image compositing, masking and reconstructing simulation videos.
- Visual interface: developed a cross-platform GUI with Tauri & Svelte for simulation setup, science communication and video viewing.
- Tech stack (core): Python, SQL, OpenCV, Tauri, Svelte
This project inspired my ongoing interest in designing and modelling adaptive systems, both biological and synthetic.
I am currently uploading the project to Github and will make the repository public once I have finished uploading it.