Chapter 1: IML intro
This chapter introduces the basic concepts of Interpretable Machine Learning. We focus on supervised learning, explain the different types of explanations, repeat the topics correlation and interaction.
-
Chapter 1.1: Introduction, Motivation, and History
Machine Learning Models are often black boxes, which are too complex to be understood by humans. The lack of explanation hurts trust and creates barriers. In this section we give some examples for the need of interpretability combined with a historical embedding.
-
Chapter 1.2: Interpretation Goals
There are different reasons why machine learning models need to be interpretable. In this section we have a look at four different interpretation goals: Discover and gain global insights; understand and control individual predictions; improve, debug and audit models; justification and fairness purposes.
-
Chapter 1.3: Dimensions of Interpretability
In this section several dimensions of interpretation are introduced: Intrinsic vs. model-agnostic methods, different types of explanations, local vs. global methods, model or learner explanations - with or without refits, and levels of interpretability.
-
Chapter 1.4: Correlation and dependencies
This section gives a short recap of different correlation measures and outlines the role of dependencies in interpretable machine learning.
-
Chapter 1.5: Interaction
If the influence of a feature on a prediction depends on the parameter value of another feature, the two features interact with each other. This section gives a mathematical definition and some explanatory examples of feature interactions.