NucleusResearch.com
10
Document Number: T147 October 2019
C O N C L U S I O N
Last year we saw deep learning projects moving en masse out of the classroom and into the
boardroom as advances in machine learning education, compute technology, cloud
infrastructure, and suitable datasets helped make deep learning commercially feasible. This
year, the trend continued with more and more companies looking to implement deep
learning at scale to solve real-world problems. For the second time in two years, the number
of deep learning projects in production more than doubled, primarily because of the same
key factors as last year
▪ Increased availability of cloud infrastructure and services from vendors like AWS to
support data-heavy, compute-intensive processes like deep learning.
▪ Advances to the state-of-the-art in deep learning with improved techniques, network
architectures, and datasets that make neural networks more accurate and capable.
▪ Continued investment in the community to share experience and enable other deep
learning researchers through online forums and documentation, open source
libraries and frameworks, and cloud offerings like pre-built models and specialized
hardware optimized for machine learning.
As the initial cost to explore deep learning decreases, we see more and more businesses
looking to join the fray. Rather than looking to re-invent the wheel, the most efficient
strategy to this end is to partner with a cloud vendor that has the infrastructure, expertise,
and additional services to bring deep learning from concept to completion. From our
analysis, we found that Amazon's reputation as the most mature and sophisticated
enterprise cloud technology provider along with its field-specific investments in machine
learning services and platform flexibility to support the customer's choice of network
architecture, development framework, or data sources make it the cloud platform of choice
for deep learning professionals.