Scalable Partial Explainability in Neural Networks via Flexible Activation Functions (Student Abstract)

Abstract

Current state-of-the-art neural network explanation methods (e.g. Saliency maps, DeepLIFT, LIME, etc.) focus more on the direct relationship between NN outputs and inputs rather than the NN structure and operations itself, hence there still exists uncertainty over the exact role played by neurons. In this paper, we propose a novel neural network structure with Kolmogorov-Arnold Superposition Theorem based topology and Gaussian Processes based flexible activation function to achieve partial explainability of the neuron inner reasoning. The model feasibility is verified in a case study on binary classification of the banknotes.

An Inquiry into Model Validity When Addressing Complex Sustainability Challenges

“…A definition of the substitution rules from the real world to the model and back greatly adds to interpretability. Many machine learning techniques are difficult to interpret [154], but attempts at explainable artificial intelligence [155] present an opportunity to improve complex systems model interpretability. Once a model has external validity, it can be used to see what would happen in the future (or could have happened in the past) if the inputs are changed, i.e., to measure the sensitivity of the output results with respect to changing input variables [154].…”