Do Explanations Explain? Model Knows Best

MCML Authors

Ashkan Khakzar

Dr.

* Former Member

→ Group Daniel Rückert
 Artificial Intelligence in Healthcare and Medicine

Nassir Navab

Prof. Dr.

Core PI

Computer Aided Medical Procedures & Augmented Reality

Abstract

It is a mystery which input features contribute to a neural network's output. Various explanation (feature attribution) methods are proposed in the literature to shed light on the problem. One peculiar observation is that these explanations (attributions) point to different features as being important. The phenomenon raises the question, which explanation to trust? We propose a framework for evaluating the explanations using the neural network model itself. The framework leverages the network to generate input features that impose a particular behavior on the output. Using the generated features, we devise controlled experimental setups to evaluate whether an explanation method conforms to an axiom. Thus we propose an empirical framework for axiomatic evaluation of explanation methods. We evaluate well-known and promising explanation solutions using the proposed framework. The framework provides a toolset to reveal properties and drawbacks within existing and future explanation solutions

inproceedings KKN+22

---

CVPR 2022

IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, LA, USA, Jun 19-24, 2022.

---

Authors

A. Khakzar • P. Khorsandi • R. Nobahari • N. Navab

---

Links

DOI GitHub

---

Research Area

 C1 | Medicine