Documentation

The Neurosci-ANN’s project develops a novel framework for improving the interpretability of artificial neural networks by combining methodologies from computational neuroscience and explainable artificial intelligence. Its central objective is to characterise how information is represented and processed within deep learning models, and how these internal representations give rise to their observed outputs. A key goal is to extend the state of the art in interpretability by revealing how activations of collections of artificial neurons in hidden layers are directly associated with the decision-making processes of modern AI systems.