Tutorial Description

This tutorial examines the synergistic relationship between explainability methods for machine learning and a significant problem related to model quality: robustness against adversarial perturbations. We begin with a broad overview of approaches to explainable AI, before narrowing our focus to post-hoc explanation methods for predictive models. We discuss perspectives on what constitutes a good explanation in various settings, with an emphasis on axiomatic justifications for various explanation methods. In doing so, we will highlight the importance of an explanation method's faithfulness to the target model, as this property allows one to distinguish between explanations that are unintelligible because of the method used to produce them, and cases where a seemingly poor explanation points to model quality issues. Next, we introduce concepts surrounding adversarial robustness, including state-of-the-art adversarial attacks as well as a range of corresponding defenses. Finally, building on the knowledge presented thus far, we present key insights from the recent literature on the connections between explainability and adversarial robustness. We show that many commonly-perceived issues in explanations are actually caused by a lack of robustness. At the same time, we show that a careful study of adversarial examples and robustness can lead to models whose explanations better appeal to human intuition and domain knowledge.

Syllabus

Introduction:

• Robustness cpatures Conceptual Soundness

• Explainability and Model Robustness

Section I: Foundation of Good Explanations

• Background of XAI Methods

• Evaluation Criteria for XAI Methods

• Demo: TruLens library for Model Explanations

Section II: Foundation of Adversarial Robustness

• Adversarial Attacks for Deep Neural Networks

• Building Robust Deep Nueral Networks

Section III: Connecting Explainability and Robustness

• Can We Trust Explanations

• Why Robust Models are More Explainable - A Feature Perspective

• Why Robust Models are More Explainable - A Geometric Perspective

• Demo: Boundary-based Explanations

Audience

The target audience of this tutorial are ML researchers, practitioners, and policy makers who are interested in explainablity and robustness in deep learning or traditional statistical ML models. As a result, basic understanding of the theories and architectures of common models, such as Convolution Neural Networks (CNN) and decision trees, are strongly recommended. Since we will include live-demos of some explanation tools, basic understanding of usage and applications of machine learning frameworks such as scikit-learn or tensorflow is recommended but not required.

Tutors

Demo

We present a new library, TruLens, containing attribution and interpretation methods for deep neural networks. To quickly play around with the TruLens library written for pytorch or Tensorflow Keras models, check out the following CoLab notebooks:

• Pytorch

• Tensorflow2

With support of Trulens, we also include the demo of

• Boundary-based Explanation.

Shared-source Libraries

• TruLens: cross-framework library for deep learning explainability.

• GloRo Nets: library for training networks that are provably robust to small-norm adversarial examples.

• Boundary Attributions: implementation of boundary-based explanations.