Industry executives and experts share their predictions for 2020. Read them in this 12th annual VMblog.com series exclusive.
By
Roger Levinson, COO, BrainChip
Artificial Intelligence, Deep Learning, and Neural Networks in 2020 Vision
What
end users, investors, and industry insiders can look forward to in 2020.
Five
predictions addressing the new and changing technologies that affect AI
applications.
1. Investment in today's Deep Learning Neural
Network technologies will plateau and shift to next generation AI solutions
Deep Learning accomplishments and advances have
played center stage in stories about Artificial Intelligence in recent years. With
the availability of vast computing resources and the evolution of GPU based
accelerators, Artificial Intelligence has soared to the forefront of industry
attention and investment, driving new use cases across nearly every end
market. Transformational solutions have been enabled for businesses in the
areas of finance, insurance, credit, security, social media, smart cities, just
to name a few. CIOs are driven to invest in AI as the next greatest thing
since cloud computing as a way to drive new levels of innovation and
competitiveness. However, this has all come at a cost in terms of physical
infrastructure and energy consumption and currently falls short of delivering
on the promise of Artificial Intelligence by providing only classification
based upon a pre-trained solution. The intelligence
in Artificial Intelligence is yet to be demonstrated. As markets
move from the data center to the edge, where data is created, a next generation
of AI solutions will be required and that is where investment focus will
shift in 2020.
2. Edge Applications will change the
requirements for AI solutions
To release the full potential of AI, Edge
applications, such as IoT, requires certain characteristics not available
in today's deep learning infrastructures. Edge applications include well
known examples such as Smart Cameras, Smart Home environments, autonomous
vehicles, robotics, drones, and many more. Edge applications come
with limited power budgets, physical space constraints, a wide variety
of use cases, the need for autonomous function and
the requirement for continued learning. As we enter 2020, the growth
in Edge applications will accelerate as solutions capable of achieving these
requirements will begin to enter the market. No longer will AI solutions be
measured by the amount of data that can be processed in isolated
data centers, but will be judged on the ability to solve challenging, real
world problems at the Edge.
3. Interest in 3rd generation Neural Networks will grow
Given the needs of Edge applications, next
generation technologies, such as Neuromorphic Computing, will emerge as
the predominant solution. At the heart of an AI solution is the neural network
which is intended to mimic the human brain. The solutions to date have shown
no real relationship to the human brain but are extensions of traditional
computing architectures. The limitations of past neural network approaches,
which fundamentally try to ‘shrink' the GPU-driven solution down to
fit at the edge, simply do not achieve the requirements. Third generation
neural networks based upon the fundamentals of spiking neural
networks (SNN) will lead the way in solutions for AI at the Edge. Successful
suppliers of the next generation solutions will provide technology which not
only achieves the future needs of Edge applications but will support much of
today's already established, infrastructure and ecosystem. The ability to
merge today's solutions with next generation capabilities will accelerate the
proliferation of intelligent devices and will bring intelligence to artificial
intelligence at the Edge.
4. Proliferation of Smart Edge Devices
2020 will bring a surge in Smart Devices for Edge
applications. Utilizing today's technology, early Smart Devices will continue
to proliferate, leveraging deep learning infrastructure and currently available
technologies. Applications such as Smart Speakers, Video Doorbells, remote
controls, hand-held devices and many more will continue to incrementally
improve on the AI features in their offering. However, 2020 will be a turning
point where a new class of these devices will begin to emerge with new
capabilities never seen before. Individuals will be able to personalize their
devices through human interfaces such as voice and gesture without the need for
connectivity to the Internet. Smart Cameras will be able to identify
specific individuals unique to each camera. This next step in the artificial
intelligence evolution will represent a turning point in how AI systems behave.
At the heart of this change will be Neuromorphic Computing transforming the
landscape.
5. Advances in real-time learning will take root
in 2020
The structure of the brain contributes to another
important factor. As we walk down stairs or listen to music, the brain predicts
what happens next before sensory perception arrives. Walking down the stairs
our foot ‘knows' the location of the next step. Listening to music we ‘know'
what note follows next. This is because the brain predicts the next event a few
milliseconds before it happens. When this same prediction system is used in
artificial neural networks we will start to see better learning, and the system
will learn better predictions. As opposed to today's deep learning approach
of back-propagation, continuous learning with forward-propagation, reinforced
through Spike-timing-dependent-plasticity (STDP) will provide the solution.
Artificial Neural Networks will become increasingly capable, yielding better
accuracy in predicting events, advanced episodic memory, and real-time
learning. Real-time learning of complex, real-world events is the key to
adaptability, and eventually, intelligent machines. As we continue to use SNNs,
and we start to implement these new structures, AI will no longer be restricted
to simple classification tasks.
The era of artificial intelligence has just
begun and the next decade is destined to bring dramatic and transformative change.
##
About the Author
Roger Levinson, BrainChip COO, Mr. Levinson most
recently served as Vice President of Data Management at Rstor where he
previously served as Vice President of ASIC Engineering. Mr. Levinson has also
served as Vice President of Engineer at Rambus and Vice President/General
Manager of Strategy and Innovation at Semtech. He held a variety of senior
engineering positions prior to joining Semtech including at Intersil, Xicor,
Analog Integration Partners, Exar ,and Micropower. Mr. Levison earned his
Bachelor's Degree from the University of California, Davis in Electrical and
Electronic Engineering, and also earned his Masters Degree from the University
of California, Davis in Electrical and Electronic Engineering.
(https://www.brainchipinc.com)