The Elephant in your Dataset: Addressing Bias in Machine Learning

Thanks to third-party libraries and packages, machine learning has become more accessible than ever, making data science available at your fingertips. However, as we move forward in our craft, it is crucial that we address the elephant in the room (or in the dataset): bias. Bias crops up everywhere, from the cognitive and historical biases that exist in society- to those that can be introduced in algorithms and machine learning processes.

This session is meant to encourage participants to consider their own biases (and the impact they may have on machine learning models), and to provide the tools needed to create more fair and just systems. We will begin with an interactive and human-focused discourse about bias: what it is, the forms it appears in, and how it permeates our society. Then we'll explore different types of bias within the context of machine learning using real-world stories and examples. Together we will examine the consequences of bias left unattended, including the potential for discrimination and unfair decision-making.

We will delve into the latest research and several approaches to mitigating bias, including methods in pre-processing, in-processing, and post-processing. We will also work through one example together to give participants a hands-on understanding of how to apply these techniques in practice. By understanding bias and taking the steps necessary to quell its effects, we can ensure that machine learning is used responsibly and ethically in the future, resulting in a world with more trustworthy technology.