Obviously, Father Christmas – or Santa Claus or whatever else he’s known as – is real. Whoever tells you otherwise is either misinformed or simply a liar. However, he can only deliver all of the presents for children around the world in just one evening with the help of his magical reindeers – Dasher, Dancer, Prancer, Vixen, Comet, Cupid, Donner, Blitzen and, of course, Rudolph. Unfortunately, Rudolph and his friends are susceptible to some terrible accidents. In Norway – the home of reindeer – eight train incidents killed more than 100 reindeer in a single week in 2017. The main problem is that train engineers are often unaware that reindeer herds from nearby ranches are on the trainlines and that slow train speeds are therefore necessary.
In an attempt to stop this from happening in the future (and to also save Christmas), AI tech companies Boulder AI and ENGINE27 worked together to create an Arm-powered smart camera system called the DNNCam – also known as “Reindeer Cam”. This innovative device is capable of detecting herds of reindeer through machine learning (ML) algorithms as they approach the train tracks. The Reindeer Cam then warns train operators to reduce speeds when the animals are present, with the communication taking place via satellite or cellular modem. The oncoming train slows down or stops, preventing incidents on trainlines in remote parts of Norway (close to Lapland) where reindeer are often needlessly killed.
A central feature of the Reindeer Cam is the onboard data processing capability, which is a vital part of ML at the edge. This means having the capability and flexibility to perform tasks on the device rather than on the cloud. The power and cost required to shift massive amounts of data between devices and the cloud can produce a noticeable lag or delay, but this is avoided with on-device processing within the device. As you can imagine, any lag or delay in sending information about reindeer being on the train tracks could lead to dire consequences, so time is of the essence.
The Reindeer Cam has an NVIDIA TX2 GPU module onboard, which enables the ML compute performance at the edge. The TX2 features a hex-core Armv8 64-bit CPU complex, which combines a dual-core NVIDIA Denver 2 alongside a quad-core Arm Cortex-A57 CPU. Arm’s Cortex-A processor series is designed for devices undertaking complex compute tasks, with the Cortex-A57 CPU supporting a wide range of applications that require high-performance processing combined with power efficiency. Since the Cortex-A57, we have continued to make ML improvements to our Cortex A-processor range. In fact, the new Cortex-A76 CPU delivers 4x compute performance improvements for ML at the edge, which enables responsive and secure experiences on a range of devices from cameras to smartphones.
Similar to the Reindeer Cam, there are already plenty of ‘mission-critical’ applications that use ML at the edge, particularly those in healthcare, security, agriculture and environmental monitoring. Going back four weeks, Thanksgiving took place in the U.S., which traditionally marks the end of the harvest season in the country. During the intensive harvest period, farmers will work around the clock to ensure that all the crops are gathered and ready to be sent on to supermarkets and stores. However, a big challenge is making sure that ‘bad crops’ are not harvested as part of the process. Luckily, ML on mobile is already helping farmers overcome this particular challenge.
A smartphone-based program has been developed that uses ML to automatically detect diseases in the cassava plant – the most widely grown root crop in the world – with nearly 100 percent accuracy. The technology is a great example of ML at the edge, as it uses the neural network that powers the ML entirely on the Arm-powered smartphone device, so no cloud computing is required. The app provides farmers with a tool to weed out diseased cassava plants in an inexpensive, easy and accurate way. It has already provided significant costs and time savings through identifying diseased crops that farmers would not have been able to spot without the app which is deployed on Arm-based mobile devices.
Whether it’s helping to save reindeer or crops, ML at the edge is being used across a range of transformative applications and technologies. As I noted in an article for AI Business, ML at the edge is only getting started, but there are already a range of positive use cases showing it in action. The example of the Reindeer Cam is by far one of the most positive, particularly as we approach Christmas Day and Father Christmas’s busy schedule on Christmas Eve! In this regard, I think one could say that ML is actually helping to ensure that Christmas runs as smoothly as possible. And long may it continue in the future! Have a Merry ML Christmas!
[CTAToken URL = "https://www.arm.com/products/silicon-ip-cpu/machine-learning/project-trillium" target="_blank" text="Learn more about ML at Arm" class ="green"]