Machine learning tackles the everyday and makes our lives easier
A New Zealand startup that makes its own servers is expanding into the artificial intelligence space, developing machine learning solutions that perform common tasks while relieving people of repetitive, unsatisfying work. More recently, having recognized an opportunity for developing low-cost, highly efficient, and environmentally responsible hardware, Kauricone has veered in an intriguing direction: developing software that thinks about everyday problems so we don’t have to. These tasks include identifying litter for improved recycling, “looking” at items on streets for automated safety, identifying vermin, and—as the ultimate facilitation of a notoriously sleep-inducing task—counting sheep.
According to Mike Milne, CEO, founder and technology industry veteran, Kauricone products include application servers, cluster servers and Internet of Things servers. It was in this latter category that the idea of applying machine learning to the edge of the network emerged.
“Having already developed low-cost, low-power edge hardware, we realized that there was a huge opportunity for applying smart computing to some downright inconvenient everyday tasks,” said Milne. “After all, we already had all the basic building blocks: the hardware, the programming ability and, with good cell phone coverage, the connectivity as well.”
Work is just another name for tasks that people would rather not do themselves or that we cannot do ourselves. And despite living at fabulously advanced ages, there’s a persistent reality of all sorts of tasks that need to be done every day that don’t require a particularly high level of commitment or even intelligence.
Machine learning (ML) is often a promising solution for these tasks in particular. “ML collects and analyzes data using statistical analysis and pattern matching to learn from past experiences. With the trained data, it gives reliable results and people can stop doing the tedious work,” says Milne.
In fact, there is more to it than meets the eye (so to speak) in terms of computer image recognition. Therefore, “capcha” challenges are often little more than “identify all images that contain traffic lights” because distinguishing objects is difficult for bots. ML overcomes the challenge with the “training” mentioned by Milne: the computer is shown thousands of images and learns which are hits and which are misses.
“There may be as many use cases as there are boring but necessary tasks in the world,” notes Milne. “So far we have tackled a few. Rocks on roads are dangerous, but monitoring thousands of miles of asphalt comes at a price. Construction waste is extensive, harmful to the environment and should be better disposed of. Sheep are plentiful and not always in the right paddock. And pests are a threat to New Zealand’s biodiversity.”
To address each of these issues, Kauricone started with its homegrown RISC IoT server hardware as a base. The servers run on Ubuntu and are programmed with Python or other open source languages. They typically come with 4GB of RAM and 128GB of solid-state storage, the solar-powered Edge devices draw just 3 watts and run indefinitely on a single solar panel. This makes for a reliable, cost-effective, turnkey device, says Milne.
The Rocks on Roads project illustrated the challenges of “simple” image identification, with Kauricone eventually running a training model 24/7 for 8 days, collecting 35,000 iterations of rock images that expanded to 3,000,000 identifiable features (remember, a Humans recognize a stone almost immediately, perhaps more quickly if thrown). With this training, the machine became very good at spotting rocks on the roads.
For a new construction waste project, the Kauricone IoT server will closely monitor the type and amount of waste entering the construction site dumps. The resulting data, trained to identify waste types, provides the basis for improving waste management and recycling, or redirecting certain items for more responsible disposal.
Counting sheep isn’t just a way to reduce bedtime, it’s also an essential chore for farmers across New Zealand. That’s not all – as an ML exercise, it anticipates the potential for smarter inventory management, as does the pest identification test case pursued by Kauricone. The ever-vigilant camera and supporting hardware perform multiple tasks: identifying individual animals, numbering them, and monitoring grass stock, which is essential for sheep nutrition. This application has so far been tested on a small herd and is ready for scale-up.
According to Milne, the small test cases that Kauricone has been pursuing so far are just the beginning and anticipate significant potential for ML applications in all walks of life. “The number of daily tasks where computer vision and ML can reduce our workload and contribute to improved efficiency and ultimately a better and more sustainable planet is literally endless,” he notes.
The Rocks on Roads project promises improved safety with less “human” overhead, reducing or eliminating the possibility of human error. Waste management is a complex problem in which simple economics (and potentially dumbing work) make it difficult to employ staff; New Zealand’s primary sector is ripe for technology-driven performance improvements that could boost already impressive productivity through automation and improved control; and pest control can help the Department of Conservation and allied parties achieve better results with fewer resources.
“It’s still early days,” says Milne, “but the results of these exploration projects are encouraging. With the connectivity of constantly growing cellular and low-power networks such as SIGFOX and LoraWan, the corresponding infrastructure is increasingly available even in remote locations. And purpose-built low-power hardware pushes computing power to the edge. Now it’s just a matter of identifying opportunities and building the applications.”
Visit Kauricone’s website for more information.
#Machine #learning #tackles #everyday #lives #easier Source