Stephanie Eckman
SoDa Symposium: Human-Centered Data Collection for Quality AI: Lessons from Survey Research
A Presentation with Q&A
presented on
Tuesday, December 16, 2025
12 pm – 1 pm (Online)
Abstract:
Machine learning model performance depends critically on training data quality, yet many researchers lack systematic approaches to human data collection. This presentation highlights findings from a series of experiments applying established principles from survey methodology to improve data labeling practices in computational research. Stephanie Eckman will demonstrate how labeler characteristics, task design, and sampling strategies affect data quality and ultimately robustness, fairness, and generalizability.
Presenter:
Stephanie Eckman, Ph.D.
Principal Research Scientist, Amazon
Researcher and Data Scientist, Social Data Science Center, University of Maryland
Bio:
Stephanie Eckman has a PhD in Methodology and Statistics from the University of Maryland. She has collected survey data around the world for government, nonprofits, and industry. Her current research interest is in applying the lessons from surveys to collect more accurate and efficient training data for AI/ML models. Her work on this topic has been published at EMNLP and ICML.
Moderator:
Frauke Kreuter
Professor
Joint Program in Survey Methodology
Co-Director, Social Data Science Center (SoDa)
University of Maryland
Chair of Statistics and Data Science in Social Sciences and Humanities
Ludwig-Maximilian, University of Munich
