What was once considered something out of a science-fiction movie is now being applied to almost every industry you can think of. Artificial intelligence and deep learning are here and they are increasing efficiency across industries.
Over the past few years, the artificial intelligence (AI) market has grown rapidly. With the advancements in processing power and the plethora of information now available as a result of IoT proliferation, more and more industries are putting AI to work. In addition to AI, deep learning has also gained prominence. Despite often being lumped together with machine learning, deep learning takes machine learning a step further by creating smarter, more actionable intelligence through the use of deep neural networks (DNN).
In markets such as retail, an increasing necessity for superior surveillance and monitoring at physical stores, growing awareness and application of AI in the retail industry, enhanced user experience, improved productivity, return on investment (ROI), mainlining inventory accuracy, and supply chain optimization are all driving AI adoption, according to a report by MarketsandMarkets. Growth of AI in the global retail market is expected to reach US$5 billion by 2022, up from $736.1 million in 2016, the report stated.
The market for deep learning is expected to be worth $1.8 billion by 2022, growing at a compound annual growth rate (CAGR) of 65.3 percent between 2016 and 2022, according to MarketsandMarkets. The report attributes growth to increased R&D activity for the development of better processing hardware for deep learning, growing adoption of cloud-based technology, and the growing usage of the deep learning technology across various industries.
In the U.S. alone, the deep learning market is expected to grow at a CAGR of over 57 percent during the period 2017 to 2021, as stated in a report by Technavio. “Industries across the U.S. are striving hard to channelize and optimize multiple facets of operations, including data analysis, storage, strategy and decisionmaking. Deep learning has surfaced as a powerful tool that assists industries in improving the programming of automated machines/equipment and inducing self-learning capabilities,” the report said.
Tom Edlund, CTO,
BriefCam
Not Without Challenges
Deep learning and AI are faced with many challenges when it comes to application. “Our two main challenges with deep learning are: training data — generating massive amounts of accurately annotated data for training requires large amount of resources. We overcome this by developing advanced tools for semi-automated training data generation. And hardware costs — running deep learning (inference) is expensive, particularly on servers where energy efficiency is critical. We combine deep learning with computer vision algorithms to keep the costs at a reasonable level,” said Tom Edlund, CTO of
BriefCam.
Jonathan Chu, CTO of
Verdigris Technologies, pointed out that it can also be challenging to calculate a favorable ROI in use cases such as condition-based maintenance or avoided costs to equipment failure. “Cost savings can be difficult to quantify or ongoing computational costs can render a project cost prohibitive. We present a value proposition to customers who benefit from Verdigris solutions through reduced capital and operating expenditures,” he added.
Hardware Improvements Drive Deep Learning Growth
Running deep learning algorithms requires high-performance hardware. This isn’t always available or affordable. However, advancements in CPUs and GPUs are making the use of deep learning and AI more accessible than ever.
“The emergence of GPU technology, in particular, has led to a dramatic increase in performance and value. With the democratization of video analytics and increased use of AI and deep learning, this technology provides scalable solutions that can be deployed across a range of verticals and applications to better address security challenges,” said Willem Ryan, VP of Global Marketing and Communications at
Avigilon.
Edlund pointed out that as a result of the heavy use of DNNs, BriefCam recommends all its customers use NVIDIA GPUs for video processing, Tesla GPUs for servers, and GTX GPUs for workstations and laptops.
On the other hand, Chu explained: “Our hardware solution includes sensors and modules developed in-house to collect high-fidelity data. The data is warehoused on enterprise-grade Amazon Web Services (AWS) with separate VPNs for each node through independent 4G connections. The cloud-based computational structure uses CPU and batch services in AWS. We have found using CPU to be a more cost effective computational structure than using GPU.”
Ultimately, CPU or GPU, advances in both technologies will ensure rapid growth and use of both AI and deep learning in every industry.