We close out the year with Niels Bantilan talking about how to reduce bias & discrimination that can unintentionally exist in machine learning models.

Thank you to the AWS Loft (https://aws.amazon.com/start-ups/loft/ny-loft/) for hosting us. The sign-in process can be time consuming so please allow extra time to enter the building. We will try to move people inside as quickly as possible.

About the Talk:

Machine learning (ML) has the ability to learn the discriminatory patterns encoded in training data to influence socially sensitive decisions like parole-granting, hiring, & loan-approval. Many ML applications in these areas are classifiers that are part of larger systems that have a "human in the loop" who uses the predictions generated by these classifiers to make better informed decisions.

While this goal might be well-intentioned, measures must be taken to ensure that machine learning models dont reproduce discriminatory patterns in training data because we risk amplifying those patterns in the real world. In this talk, Ill be describing my thinking around what a "fairness-aware machine learning interface" means & showing a bit of the functionality of themis-ml, a package that enables the user to measure & reduce potential discrimination in supervised ML models.

About Niels:

Niels Bantilan is a data scientist at Arena (https://arena.io (https://arena.io/)), a startup that applies machine learning to transform the labor market by finding the best fit between people & organizations, primarily in healthcare organizations. He builds models to improve outcomes like retention rate, patient satisfaction, & employee engagement, & also builds internal tools to facilitate data science R&D. His research interests include reinforcement learning, artificial neural networks, & fairness-aware machine learning. Recently, he released themis-ml ( https://github.com/cosmicBboy/themis-ml ), which is a Python library built on top of sklearn that implements an API for discrimination discovery & fairness-aware modeling.

Pizza (https://nyhackr.org/pizzapoll.html) begins at 6:15, the talk starts at 7, then after we head to the local bar.