My doctoral thesis, defended at Imperial College London (Neural Reckoning Group, supervised by Prof. Dan Goodman), explores modularity and self-organisation in neural networks.

The first half of the thesis investigates the structure–function relationship in neural networks: how structural modularity, resource constraints, and input statistics jointly shape functional specialisation, and how compositional learning can be grounded in physically embedded, energy-constrained substrates such as memristive neuromorphic hardware.

The second half steps a level deeper, into the foundations of self-organisation itself. It explores how continuous Neural Cellular Automata can be sculpted into a universal computational medium via gradient descent, and how a closely related local-message-passing policy can grow and self-repair digital Boolean circuits — bridging biological resilience and reconfigurable hardware.

Together, these chapters frame modularity less as a fixed architectural property and more as an emergent phenomenon arising from the tension between constraints and goals, between local rules and global behaviour.