You're reading from Interpretable Machine Learning with Python Build explainable, fair, and robust high-performance models with hands-on, real-world examples

Product type Paperback

Published in Oct 2023

Publisher Packt

ISBN-13 9781803235424

Length 606 pages

Edition 2nd Edition

Languages

Python

Tools

BERT

Concepts

Machine Learning

Author (1):

Serg Masís

Preface

1. Interpretation, Interpretability, and Explainability; and Why Does It All Matter?

2. Key Concepts of Interpretability FREE CHAPTER

3. Interpretation Challenges

4. Global Model-Agnostic Interpretation Methods

5. Local Model-Agnostic Interpretation Methods

6. Anchors and Counterfactual Explanations

7. Visualizing Convolutional Neural Networks

8. Interpreting NLP Transformers

9. Interpretation Methods for Multivariate Forecasting and Sensitivity Analysis

10. Feature Selection and Engineering for Interpretability

11. Bias Mitigation and Causal Inference Methods

12. Monotonic Constraints and Model Tuning for Interpretability

13. Adversarial Robustness

14. What’s Next for Machine Learning Interpretability?

15. Other Books You May Enjoy

16. Index

Learning about evasion attacks

There are six broad categories of adversarial attacks:

Evasion: designing an input that can cause a model to make an incorrect prediction, especially when it wouldn’t fool a human observer. It can either be targeted or untargeted, depending on the attacker’s intention to fool the model into misclassifying a specific class (targeted) or, rather, misclassifying any class (untargeted). The attack methods can be white-box if the attacker has full access to the model and its training dataset, or black-box with only inference access. Gray-box sits in the middle. Black-box is always model-agnostic, whereas white and gray-box methods might be.
Poisoning: injecting faulty training data or parameters into a model can come in many forms, depending on the attacker’s capabilities and access. For instance, for systems with user-generated data, the attacker may be capable of adding faulty data or labels. If they have more access...

The rest of the chapter is locked