In case you missed it, we recently held a webinar on the ArrayFire GPU Computing Library and its applications to Machine Learning on June 15.

This webinar was part of a free series of webinars that help you learn about ArrayFire and Jacket (our MATLAB® product). Anyone can attend these webinars, for they are absolutely free and open for anyone to attend and interact with AccelerEyes engineers. Learn more at http://www.accelereyes.com/webinars.

Chris, a Software Engineer at AccelerEyes, explained ArrayFire’s position in the GPU computing world, and presented benchmarks where ArrayFire beats GPU libraries such as Thrust in many critical applications. He also mentioned that ArrayFire could be used either standalone, or in combination with other options for GPU computing such as OpenACC, raw CUDA or OpenCL.

Chris then proceeded to walk through an example of neural network training on sample images. This example is packaged with ArrayFire and can also be accessed in ArrayFire documentation.

The next example of machine learning with ArrayFire was the K-means algorithm, which is a popular starting point for many machine learning problems. This example can be accessed in the ArrayFire package, and the source code is available here.

A popular question we encounter (especially recently) is whether Genetic algorithms can be GPU-accelerated, and the answer is Yes! This class of algorithms use techniques inspired from natural evolution to generate useful solutions to optimization and search problems, and they are well-suited to GPU computing. Some example genetic algorithm code for ArrayFire can be viewed here.

Principal component analysis, also called the KL Transform or Hotelling Transform, is another popular algorithm used for creating predictive models. This example can be viewed here. Note that the Linear algebra package is required to run Principal Components Analysis – read the page on ArrayFire licenses.

All the examples that Chris discussed, and many many more, are freely available at the Examples page. ArrayFire’s single-GPU version is *totally free, no-obligation *and has absolutely no performance throttling.

Here’s the video with detailed discussions on each example: