Notes from Wednesday 10/16.

Link to the full textbook here.

Outline of Topics

• Introduction
• Train / Test Mismatch
• Unsupervised Adaptation
• Supervised Adaptation
• Fairness and Data Bias
• How badly can it go?
• Vocabulary

Introduction

Machine learning algorithms must make predictions based on the "right" assumptions of a particular training set, as opposed to its idiosyncrasies.

→ these idiosyncrasies are called artifacts.

Through this chapter we will look at two cases of bias negatively affecting output on test data:

1. the effect of gender in speech recognition systems
2. the effect of race in predicting criminal recidivism (likelihood of future crimes)

Train / Test Mismatch

Referred to here as the **adaptation ****problem, the problem to solve is:

"Given training data from one "old" distribution, learn a classifier that does a good job on another related, but different, "new" distribution."