NNVisualiser: An Emerging Framework for Visualising Artificial Neural Networks

NNVisualiser is a novel and nascent framework, designed for visualising the functional transformations of input in neurons of Artificial Neural Networks (ANNs). This paper presents the NNVisualiser framework implemented as a Python package, which facilitates visual understanding of the transformation of input data through various neural network layers, providing insights into network behaviour and aiding in model interpretation. The framework's capabilities include plotting inputs, scales, translations, activations, and flow diagrams of neuron activities. The framework currently supports visualising ANN models trained by Keras deep learning framework, specifically those with single hidden layers and one-dimensional continuous inputs and outputs. Future expansions of NNVisualiser will include support for multidimensional networks and models generated from several deep learning frameworks.