Outcomes, Insights and Best Practices from IIC Testbeds: Deep Learning Facility Tesbed
attributed to knowledge gained from the
IISF.
T ESTBED P LANNING
Prior to the testbed, Dell and Toshiba had
formed the basis of a relationship. The IIC
helped to formalize this relationship into a
partnership as it has done with other testbed
teams. SAS was introduced to the team
through the IIC ecosystem. SAS had prior
knowledge in deep learning and AI analytics,
as well as a campus with multiple facilities.
Toshiba and Dell needed access to a singular
campus with multiple buildings for Phase 2
of the testbed, and SAS was the ideal fit.
The Industrial Internet of Things Analytics
Framework (Industrial IoT Analytics
Framework or IIAF) also plays a role in the
testbed. The IIAF spells out how companies
who want to implement AI for IoT can search
through different design patterns and
choose those that match their respective
cases. In this way, the IIAF has helped guide
the design of the testbed, and the testbed
provides validation of the framework by
showing what works in the field.
From SAS’s perspective, the Deep Learning
Facility Testbed represented an interesting
project that dovetailed with some of SAS’s
ongoing activity. There was an opportunity
to build upon existing commonalities,
analyze differences between facilities and
expand knowledge into areas to which each
company may have had no previous direct
connection. No single company has the
hardware, software and domain expertise
needed to implement the complete project.
The IIC ecosystem helps provide the needed
capabilities to achieve the testbed goals.
T ESTBED R ESULTS
Three primary phases were laid out for the
Deep Learning Facility Testbed:
Phase 1 is centered around the
training and analysis of AI for certain
usage scenarios in the Toshiba Smart
Community
Center—specifically
anomaly detection;
Phase 2 will implement the new
technology into various different
facilities, particularly the SAS Smart
Campus; and
Phase 3 involves moving into a public
facility.
IIC I NTERACTIONS
When the Testbed was first proposed, the
Industrial Internet Reference Architecture
(IIRA) and Industrial Internet Security
Framework (IISF) played roles in formalizing
how the system worked. As a standards-
based
architectural
template
and
methodology, the IIRA enabled the testbed
team to design their system based upon
common frameworks and concepts in AI and
deep learning technology. Similarly, the
security design perspective of the testbed is
The primary usage scenario that has been
completed for Phase 1 is the anomaly
detection of facility assets. Now the focus is
shifting toward energy efficiency and the
visitor-experience areas.
One of the major areas of experimentation
in Phase 1 is how to incorporate the
computational resources for deep-learning
activity in an intensive AI environment
comprising many data streams and sensors.
- 19 -
June 2019