A practical guide toward explainability and bias evaluation in machine learning

This will be presented in English.

Alejandro Saucedo (The Institute for Ethical AI & Machine Learning)
09:0012:30 Wednesday, June 19, 2019

必要预备知识 (Prerequisite Knowledge)

  • A basic understanding of machine learning in Python

您将学到什么 (What you'll learn)

  • Gain a high-level philosophical overview of the concept of bias in machine learning, which will remove ambiguity and help simplify the challenge when faced on a practical situation
  • Learn the three key steps to assess bias throughout the lifecycle of a machine learning model
  • Understand how key machine learning concepts, such as feature importance, class imbalance, model analysis, and partial dependence, are used in a practical example, as well as how these data science fundamentals can be used to interact with key domain experts

描述 (Description)

Numerous high-profile incidents have proved undesired bias in machine learning a worrying topic. It’s particularly challenging since it could even be said that the concept of societal bias is inherently biased in itself, depending on an individual’s (or group’s) perspective.

Alejandro Saucedo demystifies machine learning bias without reinventing the wheel, instead choosing to use traditional methods to simplify the issue so it can be tackled from a practical perspective. Alejandro discusses the high-level definitions of bias in machine learning as two constituent parts: a priori societal bias and a posteriori statistical bias. He then walks you through a hands-on example—automating the loan-approval process for a company using machine learning—using key tools and techniques from the latest research, allowing you to assess and mitigate undesired bias in machine learning models. You’ll learn a pragmatic process to assess bias in machine learning models in three key steps—data analysis, inference result analysis, and production metrics analysis. At each step, you’ll discover how undesired bias is introduced, along with interesting research findings in this topic. Spoiler alert: Alejandro takes a pragmatic approach, showing how any nontrivial system will always have an inherent bias; the objective is not to remove bias but to make sure you can get as close as possible to your objectives, and you can make sure your objectives are as close as possible to the ideal solution.

Some bias may not affect the results in a negative way, and Alejandro explores techniques you can use to ensure you perform a reasonable analysis. His objective isn’t to show how to completely remove bias from a machine learning model but instead to provide you with the tools and techniques available, as well as key touch points and metrics that ensure the right domain experts are involved. He discusses fundamental topics in data science, such as feature importance analysis, class imbalance assessment, model evaluation metrics, partial dependence, and feature correlation. More importantly, he takes you through how these fundamentals can interact at different touch points with the right domain experts to ensure undesired bias is identified and documented.

Photo of Alejandro Saucedo

Alejandro Saucedo

The Institute for Ethical AI & Machine Learning

Alejandro Saucedo is chairman at the Institute for Ethical AI & Machine Learning. In his more than 10 years of software development experience, Alejandro has held technical leadership positions across hypergrowth scale-ups and tech giants including Eigen Technologies, Bloomberg LP, and Hack Partners. Alejandro has a strong track record of building multiple departments of machine learning engineers from scratch and leading the delivery of numerous large-scale machine learning systems across the financial, insurance, legal, transport, manufacturing, and construction sectors in Europe, the US, and Latin America.