Machine Learning (ML) is an exciting new subfield of computer science. With origins in pattern recognition, today’s hardware and software advances have made ML a new tool for many types of organizations I order to remain competitive. With today’s hardware, massive amounts of data can be fed into a system, which can then use algorithms to determine possible outcomes of a task, and store that information for further use.
As the amount of data that is ingested increases, the accuracy of the outcomes can improve. Similar to simulations that can give more accurate results with faster processing, more memory and improved algorithms, so can ML applications. Accelerated hardware has much to do with the improvement of ML systems today and moving into the future.
The Intel Xeon Phi processor, together with tuned libraries, can accelerate the gains the ML applications promise. The Intel Xeon Phi processor is a multicore processor that is available as a self-boot system. This means that the entire OS and applications can live entire on the coprocessor system, without having to interact frequently with the host system. Since there is no wait time for data transfer, the Intel Xeon Phi processor can accelerate well designed applications much faster than before. With up to 72 processing cores, the Intel Xeon Phi processor x200 can accelerate applications tremendously. Each core contains two Advanced Vector Extensions, which speeds up the floating point performance. This is important for machine learning applications which in many cases use the Fused Multiply-Add (FMA) instruction.
The specs for the Intel Xeon Phi Processor x200 are readily available at http://www.intel.com/content/www/us/en/processors/xeon/xeon-phi-detail.html
Since the Intel Xeon Phi processor runs applications that may have been developed for a host processor, creating Machine Learning applications becomes much easier. The Intel Xeon Phi processor is binary compatible with popular host based CPUs such as the Intel Xeon CPU. In order to develop machine learning applications, libraries such as the Intel Math Kernel Library are easy to use and are tuned for the Intel Xeon Phi processor. Just as important, are tuning of popular machine learning frameworks such as Caffe and Theano, which are then used by application developers. Using state of the art and easy to use accelerators can make writing innovative applications easier, especially when binary compatibility is maintained.