Event Date
Speaker: Xiaoyue Li (Software Engineer, Google)
Title: Ad Click Prediction - A Look at Industrial-Scale Ads Recommendation Models
Abstract: This talk offers a practical overview of lessons learned from training and deploying click-through rate (CTR) prediction models at industrial scale. We will explore the key trade-offs between model accuracy, cost, and system complexity. The presentation will outline proven techniques for optimizing performance throughout the machine learning lifecycle, including architectural choices, effective training strategies, and approaches for addressing challenges such as reproducibility and generalization across diverse UI treatments.
Bio: Xiaoyue is a software engineer at Google, working on the development of search ads prediction models. She is an alumna of the Statistics Department at UC Davis (class of 2020). This talk explores the challenges of ad click prediction at industrial scale, highlighting the unique challenges of the problem and showcasing a selection of practical solutions being developed by her team.