Machine Learning - Research (EN)

Research Guidebook: Deep Learning on AWS

Issue link:

Contents of this Issue


Page 1 of 9 2 Document Number: T147 October 2019 T H E B O T T O M L I N E Deep learning remains one of the hottest topics in the area of artificial intelligence (AI) and is rapidly moving from academia and theory to operationalized, value-add workloads. Advances in data infrastructure and compute power, as well as new classes of neural networks have helped make deep learning feasible for modern businesses to leverage, while it was still generally a pipe dream as little as one year ago. Partnering with cloud vendors like Amazon Web Services (AWS) and taking advantage of cloud-hosted machine learning services like Amazon SageMaker has been key for companies looking to accelerate deep learning projects from concept to production. In the past where the infrastructure wasn't as advanced and machine learning services were immature or not yet available, organizations without the budgets and expertise to grow AI systems from the ground up internally were left on the outside looking in as larger companies with deeper pockets and developer benches researched and implemented these AI capabilities. To understand the state of deep learning adoption and usage today, as well as how it has changed since last year's report (Nucleus Research s180 – Guidebook: TensorFlow on AWS –November 2018), Nucleus conducted interviews and analyzed the experiences across 316 unique projects. For the second time in two years, the number of deep learning projects in production more than doubled. We found that 96 percent of deep learning is running in a cloud environment, with TensorFlow being the most popular framework, being used in 74 percent of deep learning projects. PyTorch was also used in 43 percent of projects (please note that most projects leverage multiple frameworks; MXNet, Keras, and Caffe2 also appeared, virtually always in conjunction with TensorFlow, PyTorch, or both), a significant increase in adoption from last year. Of the 316 total projects, only 9 percent were built with a singular framework. Most notably, of the cloud-hosted deep learning projects, 89 percent are running on AWS; a key driver of this is the breadth of framework choices on AWS along with its own continued investment in existing and new services. We also found that 85 percent of cloud-based TensorFlow workloads are running on AWS, and 83 percent of cloud-based PyTorch projects are on AWS. Last year, about a third of the interviewed customers were either considering or using SageMaker, Amazon's managed service for building, training, deploying, and orchestrating deep learning models at scale; of the interviewed users this year, 63 percent of Amazon customers had begun using SageMaker.

Articles in this issue

Links on this page

view archives of Machine Learning - Research (EN) - Research Guidebook: Deep Learning on AWS