Univa recently sponsored an industry-wide survey to better
understand what key challenges technology and IT professionals are currently tackling
that are preventing them from moving their machine learning (ML) projects into
production. To understand more about this survey and results, VMblog spoke with
Gary Tyreman, president and CEO of Univa.
VMblog: You recently conducted
a survey of technology and IT professionals on the topic of machine learning.
What does the survey include, and what were you looking to understand from this
survey?
Gary Tyreman: Yes, we sponsored a survey, which was conducted by Dimensional
Research, that polled 344 technology and IT professionals across the globe and
across 17 industries. Given our existing work with extreme-scale, machine
learning customers like Tusbame3 and ABCI, who are running some of the largest
NVIDIA and machine learning clusters in the world, we wanted to verify if the
anecdotal evidence we are receiving from our customers is in line with the
results from this survey.
VMblog: Where there
any findings that surprised you?
Tyreman: Given our existing experience working with
customers who are running some of the world's largest machine learning
clusters, it was a pleasant surprise to see such a direct correlation between
HPC and ML, with more than 88% of respondents indicating that they are working
with HPC in their jobs. That said, we were not at all shocked that nearly 9 out
of 10 companies surveyed expect to use GPUs as part of their ML infrastructure,
since this is very much aligned with what we have witnessed with our own customer
base. Our team has seen GPUs as drivers for moving HPC workloads to
the cloud to access expensive resources. It was very inspiring for our team to
see these results.
VMblog: What were the
key findings from this survey?
Tyreman: Some of the key takeaways from this survey include
the following:
- Nearly every company surveyed (96%) stated that ML
projects will grow over the next two years, signifying incredible momentum in
this space.
- More than 80% of respondents plan to use hybrid
cloud and GPUs for ML projects.
- Though 69% of companies surveyed have three or more
teams requesting ML projects, only 2 in 10 companies have ML projects running
in production, citing migration of workloads as their biggest technical
challenge.
-
Of the 17 industries interviewed in this
survey, technology, financial services and healthcare are clearly leading the
charge when it comes to ML adoption.
VMblog: More than 50%
of respondents cited migration as the top technical issue. Why do you think
this is the case?
Tyreman: Yes, it was interesting to see that the migration of data
and applications were the top challenges for ML projects, as noted by
respondents in this survey. What this tells us is that HPC users are looking
for guidance and solutions that help them to fully
utilize and scale their ML projects and resources across their on-premise,
hybrid and cloud infrastructures.
VMblog: Were there
any industries that seem to be adopting ML faster?
Tyreman: Of the 17 industries represented in this survey, respondents expected
technology, financial services and
healthcare to lead the charge in ML adoption. These industries are not
surprising to us, given the potential ROI from machine learning and the extremely
large volumes of data needed for applications such as financial trading,
medical diagnoses, and fraud detection, where they are seeking to employ machine
learning and big data in their production workflows.
VMblog: What did the
results ultimately reveal to you about machine learning projects?
Tyreman: A result we expected and the responses backed
it up is that machine learning is poised for an explosive growth phase, with
nearly every company anticipating more projects over the next two years. We
were interested to see that the projects are being driven by numerous
stakeholder groups within the company, and that there is a diverse set of ML projects
that companies have initiated, showing that there are several areas where ML
can drive value. Yet the most revealing aspect of this report is the need for organizations
to fully utilize and scale their ML projects and resources across their on-premise,
hybrid and cloud infrastructures.
The full report titled "The Future of Machine Learning," can be
downloaded at the following link: http://www.univa.com/resources/univa-machine-learning-survey.php
##