Time: Wednesday 7-Oct-2020 23:30
Live in 64 days & 13:23:05
Motivation / Abstract
You've seen the news stories about a husband getting a much higher credit limit than his wife. A medical diagnostic test working better for some people than others. Certain people being approved for loans at a higher rate than others. It can be tempting to say that humans are inherently biased, but that machine learning solves these types of problems. It's just a computer, which is unbiased, looking at our factual data... right? Well, no. Unfairness can creep into our data in a variety of ways, and a machine learning model can often make the problem worse, not better! Fairness in machine learning is an important topic to understand. This topic is only getting more important as we integrate machine learning more into our world, and as we understand the impact these models are having on our day-to-day lives. So, what are we to do? Well, we need to identify when unfairness happens, then understand how we can correct it.
Our talk will cover three things: 1. There are many ways to define unfairness, so we'll agree on a definition and describe what it means. 2. We'll learn techniques for detecting whether or not unfairness is present in our application. 3. If unfairness is present, how do we fix it? We'll learn techniques for correcting for unfairness. We will focus on applications in finance: think loans, income, credit scores. However, these techniques can be applied well beyond the realm of finance, so no background in finance is required! We'd love for you to bring your own applications for us to discuss.