Skip to yearly menu bar Skip to main content


Poster

Linear-Complexity Data-Parallel Earth Mover's Distance Approximations

Kubilay Atasu · Thomas Mittelholzer

Pacific Ballroom #218

Keywords: [ Parallel and Distributed Learning ] [ Natural Language Processing ] [ Large Scale Learning and Big Data ] [ Information Retrieval ] [ Combinatorial Optimization ]


Abstract:

The Earth Mover's Distance (EMD) is a state-of-the art metric for comparing discrete probability distributions, but its high distinguishability comes at a high cost in computational complexity. Even though linear-complexity approximation algorithms have been proposed to improve its scalability, these algorithms are either limited to vector spaces with only a few dimensions or they become ineffective when the degree of overlap between the probability distributions is high. We propose novel approximation algorithms that overcome both of these limitations, yet still achieve linear time complexity. All our algorithms are data parallel, and thus, we take advantage of massively parallel computing engines, such as Graphics Processing Units (GPUs). On the popular text-based 20 Newsgroups dataset, the new algorithms are four orders of magnitude faster than a multi-threaded CPU implementation of Word Mover's Distance and match its nearest-neighbors-search accuracy. On MNIST images, the new algorithms are four orders of magnitude faster than a GPU implementation of the Sinkhorn's algorithm while offering a slightly higher nearest-neighbors-search accuracy.

Live content is unavailable. Log in and register to view live content