Visualising the Training Process of Convolutional Neural Networks for Non-Experts
As convolutional neural networks impact all of society, we created a visualization tool that helps non-experts understand how they learn. Our open-source tool uses dimensionality reduction (UMAP) to create a 2D visualisation of the CNN's activity. We enrich the plot with information such as the input images, predicted classes and prediction accuracy. By plotting this every epoch, we create a video which intuitively shows the network’s behaviour changing over time.
Michelle has finished the bachelor program of Industrial Design at the University of Twente and is now pursuing a master’s degree in Data Science. She enjoys finding structured ways to solve puzzles and other challenges. In her free time, she likes to dance, sing and bake delicious vegan cakes.