As a guest user you are not logged in or recognized by your IP address. You have
access to the Front Matter, Abstracts, Author Index, Subject Index and the full
text of Open Access publications.
The k-nearest neighbor (k-NN) search is the rudimentary procedure widely used in machine learning and data embedding techniques. Herein we present a new multi-GPU/CUDA implementation of the brute-force (BF) k-NN algorithm. We demonstrate its advantages over currently the fastest GPU/CUDA implementations of BF k-NN, e.g., [1] both in terms of computational time and memory requirements. Unlike its competitors, our code scales linearly with the number of GPUs (up to 8 units) what allows for scrutinizing much larger high-dimensional datasets. We also present a new GPU implementation of the approximate k-NN algorithm used in the LargeVis data embedding algorithm [2], which is more than two orders of magnitude faster than its original CPU version. We discuss its limitations in terms of accuracy, efficiency and usefulness in data embedding.
This website uses cookies
We use cookies to provide you with the best possible experience. They also allow us to analyze user behavior in order to constantly improve the website for you. Info about the privacy policy of IOS Press.
This website uses cookies
We use cookies to provide you with the best possible experience. They also allow us to analyze user behavior in order to constantly improve the website for you. Info about the privacy policy of IOS Press.