The role of TDA in improving the interpretability and explainability of neural network architectures and machine learning models

More resistant to noise and able to retrieve information on higher dimensional aspects than what is initially seen in the data, topological invariants (shapes) of data may be studied within the overall framework of TDA. My presentation will focus on current and future efforts in order to apply persistent homology to the problem of intracellular network categorization. Data may be partitioned into hierarchical clusters using techniques from TDA, including persistent homology. In order to propose an updated version of the CNN and evaluate its performance in comparison to the original CNN, I investigate both theoretical and practical insights from topological clustering of graphs and pictures.