Home // EXPLAINABILITY 2024, The First International Conference on Systems Explainability // View article


Explain Yourself

Authors:
Holger Ziekow
Peter Schanbacher
Valentin Göttisheim

Keywords: Explainable AI; Machine Learning; Neural Networks; Shapley value approximation.

Abstract:
This paper addresses the problem of providing fast and accurate approximations of Shapley values for neural networks by embedding the approximation directly into the network architecture. The approach is tested on a synthetic and a real world dataset. The results demonstrate that integrating Shapley value approximations into the loss function enables making a trade-off between explainability and prediction accuracy, optimizing both aspects. This method yields accurate approximations while improving the model's explainability, making it more stable and easier to explain in practical applications.

Pages: 5 to 10

Copyright: Copyright (c) IARIA, 2024

Publication date: November 17, 2024

Published in: conference

ISBN: 978-1-68558-215-9

Location: Valencia, Spain

Dates: from November 17, 2024 to November 21, 2024