EAGER: Preliminary Study of Hashing Algorithms for Large-Scale Learning

Cornell University, Ithaca NY

Investigators

Abstract

Many emerging applications of data mining call for techniques that can deal with data instances with millions, if not billions of dimensions. Hence, there is a need for effective approaches to dealing with extremely high dimensional data sets. This project focuses on a class of novel theoretically well-founded hashing algorithms that allow high dimensional data to be encoded in a form that can be efficiently processed by standard machine learning algorithms. Specifically, it explores: One-permutation hashing, to dramatically reduce the computational and energy cost of hashing; Sparsity-preserving hashing, to take advantage of data sparsity for efficient data storage and improved generalization; Application of the new hashing techniques with standard algorithms for learning "linear" separators in high dimensional spaces. The success of this EAGER project could lay the foundations of a longer-term research agenda by the PI and other investigators focused on developing effective methods for building predictive models from extremely high dimensional data using "standard" machine learning algorithms. Broader Impacts: Effective approaches to building predictive models from extremely high dimensional data can impact many areas of science that rely on machine learning as the primary methodology for knowledge acquisition from data. The PI's education and outreach efforts aim to broaden the participation of women and underrepresented groups. The publications, software, and datasets resulting from the project will be freely disseminated to the larger scientific community.