Machine Learning - Research (EN)

NucleusResearch.com 5 Document Number: T147 October 2019 ▪ Ability to leverage supplemental platform capabilities and tools. Security and availability were the most commonly mentioned aspects. Properly configured cloud systems benefit from the security investments of both the customer and the cloud provider. Additionally, the ability to run models on local data centers to stay in compliance with data protection laws like GDPR is important. Along with storage and compute, cloud vendors offer tools and platform capabilities to improve the developer experience. Tools like Amazon SageMaker provide great value for the cloud customers with fully-managed end-to-end machine learning workflow - from cleaning the data to training, building and deploying models. DEEP LEARNING IS REAL Last year demonstrated a step forward for the state of machine learning with 14 percent of projects in production. We saw a similar leap forward in 2019 with 38 percent of the 316 deep learning projects being classified as in production. 89 percent of deep learning projects in production are running on AWS. Seventy six percent of the projects in production leverage TensorFlow and 28 percent of projects use PyTorch. Keras and Apache MXNet were also seen in production settings as most projects have components built with multiple frameworks. Only 9 percent of projects were built with just one framework. As companies recognized that deep learning and other AI capabilities are reaching a level of maturity that allows them to deliver legitimate business value, they've scrambled to implement "low-hanging fruit," or simpler use cases that are well-demonstrated and quick to implement. Common examples include voice interfaces for websites and applications and recommendation engines for online shopping sites. Companies are still exploring more complex deep learning applications as well, but many are still in testing. The results of this year's study suggests that companies expanded their overall deep learning investments from last year, continuing progress on more complicated, multi-year aspirational projects while adding quickly-implemented, value-add applications like recommendations, sentiment analysis in chat bots, and voice interfaces, to stay abreast of market trends and demonstrate the continued value and viability of deep learning in real- world usage. D E E P L E A R N I N G O N A W S Nucleus found that the primary reasons for choosing AWS—the breadth of platform capabilities, the relationship with Amazon, and AWS' continued investment in deep learning services—remain unchanged since last year.

• Cover