The Internet of Things is an interesting phenomenon. Most people involved in technology seem to believe that the IoT growing fast, and that it represents a big opportunity. But most people also find it difficult to say exactly what the Internet of Things actually is, and how, ultimately, it will deliver substantial value for consumers, businesses and other organizations.
I believe that the value of the Internet of Things is that it’s a way to tap into vast amounts of data from the physical world, and extract insights from that data.
My Nest thermostat is a great example. Like a traditional mechanical thermostat, it measures temperature and controls heating, ventilating and air conditioning (HVAC) equipment. But it also senses when people are present, learns my preferences, and detects when HVAC equipment is malfunctioning. It generates a monthly email report comparing my home’s recent energy use to that of prior months and identifying reasons for changes. In other words, Nest gathers data, extracts insights from that data, and uses those insights to create value – in my case, more awareness of my home’s energy use, which motivates me to be more frugal, and alerts about emerging problems with my HVAC equipment, which help me improve efficiency and avoid sudden failures.
We live in technological era. But when you stop to think about it, it’s clear that our current technology is barely scratching the surface in terms of the insights that can be gained from real-world data.
At the recent Embedded Vision Summit, Stefan Heck, CEO of the start-up Nauto, explained how his company is creating a new class of add-on devices for vehicles (a preview version of the video of Heck's talk follows). Among other things, the Nauto device allows vehicle fleet operators (for example, delivery services and utility companies) to track “near misses” – cases in which an accident was narrowly avoided – so that they can identify drivers who may be at higher risk of accidents and take remedial action before those accidents happen.
Another great example is Compology, a start-up that’s bringing the IoT to the large waste receptacles found outside of every restaurant, store and office. By monitoring how full a receptacle is and transmitting that information to the cloud, Compology enables waste hauling companies to significantly reduce costs, optimizing their truck routes and schedules so that containers are emptied when they need to be emptied, and not more often.
Nest, Nauto and Complogy are three examples of what I believe is the key to delivering value through the Internet of Things: turning real-world data into useful insights. In the next decade, I believe we’ll see thousands of new products like these deployed, from smart street lights that guide drivers to open parking spots (and perhaps send tickets to those who overstay their welcome) to mobile apps that detect when a person’s gait has changed in a way that may indicate a health problem.
The essential technological ingredients for these products are inexpensive hardware (especially sensors, processors and wireless modems), ubiquitous wireless connectivity, and reliable algorithms that extract insights from sensor data.
The hardware and connectivity challenges have been met, thanks to improvements in digital chips and deployment of wireless networks, which now make it possible to easily build small, inexpensive, low-power connected devices. The algorithm challenge, in contrast, has remained a difficult one. Developing algorithms to extract meaning from noisy, real-world data is challenging, requiring big investments and specialized skills. Take Mobileye, for example. The company is a leader in computer-vision-based automotive safety systems, and has invested on the order of 1,000 man-years in algorithm development.
Very few companies (or product development projects) can afford an investment of this magnitude (or even 1% of it) – even if they could find engineers with the necessary skills, which is often difficult. And few companies have the patience to spend years developing an algorithm. So, reducing the effort required for algorithm development is critical for the success of the Internet of Things.
Fortunately, the current generation of artificial neural networks (often called “deep neural networks”) is emerging as a powerful way to create algorithms that extract insights from messy, real-world data. Whether it’s recognizing objects in images, extracting words from speech, or detecting the condition of a road surface from a microphone in a car, deep learning techniques are finding widespread use.
One of the most interesting aspects of deep neural networks is that they are trained more than they are designed. They are generalized learning machines – or “universal approximators” – that learn through being exposed to numerous examples. This means that access to training data – and the skills to use it – is beginning to take the place of traditional algorithm-design skills. It also means that now, for the first time, we have a single set of algorithmic techniques that we can use to solve a diverse range of unrelated problems – from speech recognition to image recognition.
The more engineers that understand deep learning techniques, the more quickly we’ll be able to create new types of IoT devices with the ability to extract valuable insights from data. To help speed this process along, I’m excited to be collaborating with the primary developers of the popular Caffe deep learning framework to present a full-day tutorial on September 22 in Cambridge, Massachusetts. This tutorial will provide an introduction to deep neural networks and a hands-on introduction to the Caffe framework. For details about this unique event, please visit the tutorial web page.
Jeff Bier is president of BDTI and founder of the Embedded Vision Alliance. Post a comment here or send him your feedback at http://www.BDTI.com/Contact.
Add new comment