2025-11-08 –, Room 145
As AI increasingly powers critical systems across industries, the quality and neutrality of training data have become central to model performance — and to model risk. This talk examines how biased datasets, often stemming from historical imbalances or sampling errors, propagate through machine learning pipelines and influence outcomes at scale. We’ll explore technical pathways through which bias infiltrates — from data labeling and feature selection to model optimization — and demonstrate how even small biases can magnify under automation. Drawing from real-world case studies, we’ll discuss frameworks for bias detection, debiasing techniques, and evaluation methodologies to build more robust, fair, and accountable AI systems.
Autumn is a product manager at Microsoft Azure specializing in Linux security. In her previous role at AWS as a software engineer, she focused on the development and release of Amazon Corretto (Java) while actively engaging in the OpenJDK community; before that, she worked as an AWS NoSQL Solutions Architect and created educational content in Python and Java.
Autumn co-hosts the exciting new "Fork Around and Find Out" podcast, sharing stories on tech lessons learned, with her previous co-host of the popular "Ship It!" podcast. A proud mom and "Rewriting the Code" alumni, Autumn serves as the Board Chair of Education at MilSpouse Coders, leading the chapter in the Greater Seattle Area, due to her advocacy for collaborative learning