Vaishak Belle

Explainability in Machine Learning: A basic introduction for the uninitiated

Tutorial to be held at AAMAS-2022

Recorded videos

Part 1:

Part 2:

Part 3:


Explainability and interpretability has received considerable attention in almost all major AI conferences. However, the state of the art that is being published is somewhat inaccessible to most non-ML community members. How can the uninitiated AI researcher make sense and get herself acquainted with the literature?

In this tutorial, we provide an introduction to the fundamentals and taxonomy of ML explainability. We emphasize the interactive nature of explanations, discuss two popular techniques in detail and conclude with challenges to the area of explainability.


Artificial Intelligence (AI) provides many opportunities to improve private and public life. Discovering patterns and structures in large troves of data in an automated manner is a core component of data science, and currently drives applications in diverse areas such as computational biology, law and finance. However, such a highly positive impact is coupled with significant challenges: how do we understand the decisions suggested by these systems in order that we can trust them? In this short course, we focus specifically on data-driven methods – machine learning (ML) and pattern recognition models in particular – so as to survey and distill the results and observations from the literature.

The purpose of this course can be especially appreciated by noting that ML models are increasingly deployed in a wide range of applications. However, with the increasing prevalence and complexity of methods, business stakeholders in the very least have a growing number of concerns about the drawbacks of models, data-specific biases, and so on. Analogously, data science practitioners are often not aware about approaches emerging from the academic literature, or may struggle to appreciate the differences between different methods, so end up using industry standards such as SHAP. Here, we have undertaken a survey to help practitioners and data scientists more broadly understand the field of explainable machine learning better and apply the right tools. Our material builds a narrative around a putative data scientist, and discuss how she might go about explaining her models by asking the right questions. After briefly orienting the audience around a taxonomic and narrative-based layout of explainability techniques, we provide the key technical ideas behind 2 or 3 major techniques (SHAP, Counterfactuals and Deletion Diagnostics), and conclude with possible directions for future development. We will also provide examples of how these techniques are now being used in computer vision.

Course Contents

Motivation & background (cf. paper), explainability frameworks, 6 popular techniques around a stakeholder narrative (SHAP, Coun- terfactuals, Deletion Diagnostics, PDP/ICE, Anchors, Intrees) but 2-3 discussed in detail in the interest of time, future directions and challenges to explainability.

Who is this for

The course looks at more fundamental issues, and best benefits AI scientists, data scientists, undergraduate and graduate students who are not very familiar with state-of-the-art in ML explanability. Although we do mention and discuss state-of-the-art, material is focused on understanding key underlying issues and establishing basics.

As a full day event, we will also look to provide a very short “fundamentals of machine learning” brief before delving into motivations and techniques.

For a quick preview, see a brief at the AIUK conference, starting at 35 min

Course Delivery Information

AAMAS-2022 tutorial dates are May 9-10, 2022

Duration: full day


Vaishak Belle