Amazon has announced that it has chosen MXNet as its deep learning framework of choice for its web services(AWS). Amazon extensively uses machine learning in areas like fraud detection, abusive review detection, and book classification. Amazon also uses it in application areas such as text and speech recognition, autonomous drones etc…
Amazon did not opt for some of the popular machine learning frameworks such as TensorFlow and Torch because they stated that MXNet scales and runs better than any other framework which is out there. The notable features of the MXNet were stated as its cross-platform portability and compact size. Amazon also stated that they will be further working with other organizations to advance the MXNet.
How these frameworks are selected?
According to the blog post by Amazon, the deep learning framework is selected depending on three major factors: Ability to scale, Development speed, and Portability.
The main reason for Amazon to choose MXNet seems to be for its scalability. Werner Vogels, chief technology officer and Vice President of Amazon.com stated that the speedup gained by running MXNet across multiple GPUs was highly linear and according to the benchmarks conducted across 128 GPUs, it performed 109 times faster compared to the single GPU setup.