Applied Machine Learning Explainability Techniques: Make ML models explainable and trustworthy for practical applications using LIME, SHAP, and more

✍ Scribed by Aditya Bhattacharya

• Publisher: Packt Publishing
• Year: 2022
• Tongue: English
• Leaves: 306
• Category: Library

✦ Synopsis

Leverage top XAI frameworks to explain your machine learning models with ease and discover best practices and guidelines to build scalable explainable ML systems

Key Features
 
 

• Explore various explainability methods for designing robust and scalable explainable ML systems
• Use XAI frameworks such as LIME and SHAP to make ML models explainable to solve practical problems
• Design user-centric explainable ML systems using guidelines provided for industrial applications

Book Description

Explainable AI (XAI) is an emerging field that brings artificial intelligence (AI) closer to non-technical end users. XAI makes machine learning (ML) models transparent and trustworthy along with promoting AI adoption for industrial and research use cases.

Applied Machine Learning Explainability Techniques comes with a unique blend of industrial and academic research perspectives to help you acquire practical XAI skills. You'll begin by gaining a conceptual understanding of XAI and why it's so important in AI. Next, you'll get the practical experience needed to utilize XAI in AI/ML problem-solving processes using state-of-the-art methods and frameworks. Finally, you'll get the essential guidelines needed to take your XAI journey to the next level and bridge the existing gaps between AI and end users.

By the end of this ML book, you'll be equipped with best practices in the AI/ML life cycle and will be able to implement XAI methods and approaches using Python to solve industrial problems, successfully addressing key pain points encountered.

What you will learn
 
 

• Explore various explanation methods and their evaluation criteria
• Learn model explanation methods for structured and unstructured data
• Apply data-centric XAI for practical problem-solving
• Hands-on exposure to LIME, SHAP, TCAV, DALEX, ALIBI, DiCE, and others
• Discover industrial best practices for explainable ML systems
• Use user-centric XAI to bring AI closer to non-technical end users
• Address open challenges in XAI using the recommended guidelines

Who this book is for

This book is designed for scientists, researchers, engineers, architects, and managers who are actively engaged in the field of Machine Learning and related areas. In general, anyone who is interested in problem-solving using AI would be benefited from this book. The readers are recommended to have a foundational knowledge of Python, Machine Learning, Deep Learning, and Data Science. This book is ideal for readers who are working in the following roles:
 
 

• Data and AI Scientists
• AI/ML Engineers
• AI/ML Product Managers
• AI Product Owners
• AI/ML Researchers
• User experience and HCI Researchers

Table of Contents

 

1. Foundational Concepts of Explainability Techniques
2. Model Explainability Methods
3. Data-Centric Approaches
4. LIME for Model Interpretability
5. Practical Exposure to Using LIME in ML
6. Model Interpretability Using SHAP
7. Practical Exposure to Using SHAP in ML
8. Human-Friendly Explanations with TCAV
9. Other Popular XAI Frameworks
10. XAI Industry Best Practices
11. End User-Centered Artificial Intelligence

✦ Table of Contents

Cover
Title page
Copyright and Credits
Dedications
Contributors
Table of Contents
Preface
Section 1 – Conceptual Exposure
Chapter 1: Foundational Concepts of Explainability Techniques
Introduction to XAI
Understanding the key terms
Consequences of poor predictions
Summarizing the need for model explainability
Defining explanation methods and approaches
Dimensions of explainability
Addressing key questions of explainability
Understanding different types of explanation methods
Understanding the accuracy interpretability trade-off
Evaluating the quality of explainability methods
Criteria for good explainable ML systems
Auxiliary criteria of XAI for ML systems
Taxonomy of evaluation levels for explainable ML systems
Summary
References
Chapter 2: Model Explainability Methods
Technical requirements
Types of model explainability methods
Knowledge extraction methods
EDA
Result visualization methods
Using comparison analysis
Using Surrogate Explainer methods
Influence-based methods
Feature importance
Sensitivity analysis
PDPs
LRP
Representation-based explanation
VAMs
Example-based methods
CFEs in structured data
CFEs in unstructured data
Summary
References
Chapter 3: Data-Centric Approaches
Technical requirements
Introduction to data-centric XAI
Analyzing data volume
Analyzing data consistency
Analyzing data purity
Thorough data analysis and profiling process
The need for data analysis and profiling processes
Data analysis as a precautionary step
Building robust data profiles
Monitoring and anticipating drifts
Detecting drifts
Selection of statistical measures
Checking adversarial robustness
Impact of adversarial attacks
Methods to increase adversarial robustness
Evaluating adversarial robustness
Measuring data forecastability
Estimating data forecastability
Summary
References
Section 2 – Practical Problem Solving
Chapter 4: LIME for Model Interpretability
Technical requirements
Intuitive understanding of LIME
Learning interpretable data representations
Maintaining a balance in the fidelity-interpretability trade-off
Searching for local explorations
What makes LIME a good model explainer?
SP-LIME
A practical example of using LIME for classification problems
Potential pitfalls
Summary
References
Chapter 5: Practical Exposure to Using LIME in ML
Technical requirements
Using LIME on tabular data
Setting up LIME
Discussion about the dataset
Discussions about the model
Application of LIME
Explaining image classifiers with LIME
Setting up the required Python modules
Using a pre-trained TensorFlow model as our black-box model
Application of LIME Image Explainers
Using LIME on text data
Installing the required Python modules
Discussions about the dataset used for training the model
Discussions about the text classification model
Applying LIME Text Explainers
LIME for production-level systems
Summary
References
Chapter 6: Model Interpretability Using SHAP
Technical requirements
An intuitive understanding of the SHAP and Shapley values
Introduction to SHAP and Shapley values
What are Shapley values?
Shapley values in ML
The SHAP framework
Model explainability approaches using SHAP
Visualizations in SHAP
Explainers in SHAP
Using SHAP to explain regression models
Setting up SHAP
Inspecting the dataset
Training the model
Application of SHAP
Advantages and limitations of SHAP
Advantages
Limitations
Summary
References
Chapter 7: Practical Exposure to Using SHAP in ML
Technical requirements
Applying TreeExplainers to tree ensemble models
Installing the required Python modules
Discussion about the dataset
Training the model
Application of TreeExplainer in SHAP
Explaining deep learning models using DeepExplainer and GradientExplainer
GradientExplainer
Discussion on the dataset used for training the model
Using a pre-trained CNN model for this example
Application of GradientExplainer in SHAP
Exploring DeepExplainers
Application of DeepExplainer in SHAP
Model-agnostic explainability using KernelExplainer
Application of KernelExplainer in SHAP
Exploring LinearExplainer in SHAP
Application of LinearExplainer in SHAP
Explaining transformers using SHAP
Explaining transformer-based sentiment analysis models
Explaining a multi-class prediction transformer model using SHAP
Explaining zero-shot learning models using SHAP
Summary
References
Chapter 8: Human-Friendly Explanations with TCAV
Technical requirements
Understanding TCAV intuitively
What is TCAV?
Explaining with abstract concepts
Goals of TCAV
Approach of TCAV
Exploring the practical applications of TCAV
Getting started
About the data
Discussions about the deep learning model used
Model explainability using TCAV
Advantages and limitations
Advantages
Limitations
Potential applications of concept-based explanations
Summary
References
Chapter 9: Other Popular XAI Frameworks
Technical requirements
DALEX
Setting up DALEX for model explainability
Discussions about the dataset
Training the model
Model explainability using DALEX
Model-level explanations
Prediction-level explanations
Evaluating model fairness
Interactive dashboards using ARENA
Explainerdashboard
Setting up Explainerdashboard
Model explainability with Explainerdashboard
InterpretML
Supported explanation methods
Setting up InterpretML
Discussions about the dataset
Training the model
Explainability with InterpretML
ALIBI
Setting up ALIBI
Discussion about the dataset
Training the model
Model explainability with ALIBI
DiCE
CFE methods supported in DiCE
Model explainability with DiCE
ELI5
Setting up ELI5
Model explainability using ELI5
H2O AutoML explainers
Explainability with H2O explainers
Quick comparison guide
Summary
References
Section 3 – Taking XAI to the Next Level
Chapter 10: XAI Industry Best Practices
Open challenges of XAI
Guidelines for designing explainable ML systems
Adopting a data-first approach for explainability
Emphasizing IML for explainability
Emphasizing prescriptive insights for explainability
Summary
References
Chapter 11: End User-Centered Artificial Intelligence
User-centered XAI/ML systems
Different aspects of end user-centric XAI
Rapid XAI prototyping using EUCA
Efforts toward increasing user acceptance of AI/ML systems using XAI
Providing a delightful UX
Summary
References
Index
Other Books You May Enjoy