Nonparametric Methodology for Learning from People: Inference, Algorithms, and Optimality

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Learning from people represents a new and exciting paradigm for research in statistics and data science and is useful both in understanding behavioral patterns (for example in marketing) and for informing interventions (for example in education). Data from humans can also be elicited to inform various downstream tasks, with crowdsourcing being routinely used to collect data across applications spanning bioinformatics, epidemiology, computer vision, and environmental modeling. A common characteristic of such data is its scale and noisiness; having flexible and interpretable models for such data is broadly useful in downstream decision-making. The investigators aim to develop and thoroughly study flexible classes of models and methods for drawing inferences from large-scale data collected from people in a variety of such contexts. Specifically, the investigations will focus on developing three key facets of such nonparametric methodology: (a) Computationally efficient, assumption-lean methods to fit expressive models to data; (b) Equipping models to integrate application-specific information in a general-purpose fashion; and (c) Developing models and methods that accommodate dynamically varying data streams. The research formulates a host of theoretical and methodological questions whose solutions would constitute fundamental progress in nonparametric inference, while touching upon practical and timely issues such as fairness in ranking systems. This project will also provide training and research opportunities for the next generation of data scientists by encouraging them to model the entire data analysis pipeline, from data collection to inference. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.