Funded research project
My roles: Researcher, Software developer
My roles: Researcher, Software developer
The Grow-When-Required Neural Network is a neural network architecture that employs self-organized, unsupervised learning mechanisms.
Starting from only a few neurons, the network expands dynamically in response to the complexity and volume of information requiring representation.
The research project represented an industry-academia collaboration between Thales Group and the University of Bristol, with the objective of developing learning mechanisms applicable to autonomous maritime border patrol systems. The research findings have been published in the Neural Networks journal under the title "The robustness-fidelity trade-off in Grow When Required neural networks performing continuous novelty detection".
The network was deployed within a robotic agent navigating a simulated environment.
It incrementally acquired environmental feature representations and demonstrated the capability to identify novel objects.
The simulation implementation utilized the ARGoS large-scale multi-robot system simulator.
Experimental protocols employed custom configuration files to systematically vary network, robotic, and environmental parameters, ensuring precise experimental control. Experimental data was archived in CSV format and analyzed using integrated Python, Cython, and C++ processing pipelines.
