<meta name="viewport" content="width=device-width" />

<link rel="shortcut icon" href="favicon.ico" />

<script type="text/javascript" async src="fonts-min.js"></script>

<link rel="stylesheet" href="style.css" />

<meta charset="utf-8">

<meta name="description" content="One of the obstacles in accelerating sparse

coarsening and explore a technique for addressing load imbalance." />

<title>

Treelite: toolbox for decision tree deployment

</title>

<div id="container" itemscope

itemtype="http://www.schema.org/ScholarlyArticle">

<p>[<a href=".">&larr; Go back to profile</a>]</p>

<h1 itemprop="name">

Block-distributed Gradient Boosted Trees

</h1>

<p>

with

<a href="https://www.sics.se/people/theodore-vasiloudis">Theodore Vasiloudis</a> and

<a href="https://www.kth.se/profile/henbos?l=en">Henrik Boström</a>

</p>

<p>

Paper presented at

<a href="https://sigir.org/sigir2019/">ACM SIGIR</a> (2019), best short paper award

</p>

<h2>Download</h2>

<ul>

<li>Camera-ready version

[<a href="https://arxiv.org/abs/1904.10522">arXiv:1904.10522</a>]</li>

<li>ACM Digital Library

[<a href="https://dl.acm.org/citation.cfm?id=3331331">doi:10.1145/3331184.3331331</a>]

</li>

</ul>

<h2>Synopsis</h2>

<p itemprop="description">

The <strong>Gradient Boosted Tree (GBT) algorithm</strong> is one of

the most popular machine learning algorithms used in production, for

tasks that include Click-Through Rate (CTR) prediction and

learning-to-rank. To deal with the massive datasets available today,

many distributed GBT methods have been proposed. However, they all

assume a row-distributed dataset, addressing scalability only with

respect to the number of data points and not the number of features,

and increasing communication cost for high-dimensional data. In order

to allow for scalability across both the data point and feature

dimensions, and reduce communication cost, we propose

<strong>block-distributed GBTs</strong>. We achieve communication

efficiency by making full use of the data sparsity and adapting the

Quickscorer algorithm to the block-distributed setting. We evaluate our

approach using datasets with millions of features, and demonstrate that

we are able to achieve multiple orders of magnitude reduction in

communication cost for sparse data, with no loss in accuracy, while

providing a more scalable design. As a result, we are able to reduce

the training time for high-dimensional data, and allow more

cost-effective scale-out without the need for expensive network

communication.

</p>

<h2>Publication Details</h2>

<ul>

<li>

Conference Paper:<br>

Theodore Vasiloudis, Hyunsu Cho, and Henrik Boström.

&ldquo;Block-distributed Gradient Boosted Trees,&rdquo;

<em>ACM SIGIR 2019</em>, Paris, France, July 25, 2019.

</li>

</ul>

<p>[<a href=".">&larr; Go back to profile</a>]</p>

</div>