5 Places Arm-powered AI is Being Used to Create Change

Most machine learning (ML) today is processed on Arm central processing units (CPU) with 85 percent of smartphones running ML workloads on Arm CPUs and graphics processing units (GPU). This means that AI runs on Arm for 4 billion people on Earth.

To keep up with this demand, we continuously release new efficiency and power improvements to our processors that allow ML models to run on even the smallest endpoint devices and sensors. Arm combines hardware IP, software, and an AI development framework to guide designers in building the next generation of innovative, portable AI applications for the cloud, edge, and endpoint.

Because of this constant stream of innovation, today’s most disruptive organizations leverage Arm ML technologies to integrate new features quickly and easily across a wide range of use cases. Solutions equipped with intelligent vision, voice, and vibration capabilities have the power to advance entire industries. Here are just five examples of the many places Arm-powered technology is already making a difference.

AI in the lab

ML affects almost every industry and area of scientific research. For example, ML in engineering has been incredibly valuable for signal processing. These systems are designed to increase the accuracy and quality when sound, images, and other inputs are transmitted. ML algorithms make it possible to model signals, detect meaningful patterns, develop useful inferences, and make highly precise adjustments to signal output. In turn, signal processing techniques can also be used to improve the data provided to the ML-enabled device. This cuts out the noise that would otherwise be included in these inputs and allows engineers to achieve better results in the performance of IoT devices and other AI-enabled systems.

At Arm, we’ve used ML for many years to help improve the extremely complex validation of our processor products. AI is used to help us learn about coverage, get more accurate in where we target testing, and in designing more efficient and effective validation approaches.

AI in the field

Our latest chapter of Made Possible by Arm showcases the bee-saving AI innovation created by Beewise Technologies. This Israel-based startup — founded by professional beekeeper Eliyah Radzyner — created the Beehome: a smart beehive that houses 24 bee colonies. Each of these Beehomes is equipped with Arm-based solutions including Raspberry Pis and NVIDIA Jetsons, which powers climate & humidity control, autonomous swarm prevention, automated honey harvesting, real-time problem alerts, and pest control. 

With bee colonies currently collapsing at an alarming (and increasing) rate of 40 percent per year, this technology is preventing up to 80 percent of this loss, with 99.7 percent accurate ML models that help care for the bee colony day in day out and allow beekeepers to care for their hives from anywhere in the world. 

AI in the home

Arm partner Arçelik (Arch-lik) is a Turkish appliance manufacturer that is rethinking refrigeration technology with the goal of lowering energy consumption. 

Today, the fridge is the second-biggest energy consumer in the home, totaling 13 percent of all household energy used. Arçelik has developed a lightweight Reinforcement Learning algorithm that analyzes behavior locally, on-device, to adjust and minimize temperature fluctuations when doors are opened, reducing power use by 10 percent  in existing appliances. To explain the significance of this level of innovation, with widespread deployment of this Arm-powered technology across Europe alone, nine power plants could be rendered entirely useless.

AI in medicine

Medicine is one of the most significant industries in which AI and ML solutions can make an impact. Take hearing aids, for example. Simply amplifying sounds isn’t the solution. ML Algorithms balance sounds, ensuring the sound is clean and ideally tuned to the person’s unique type of hearing loss, allowing the brain to work in the most natural way possible.

In March 2020, as the COVID-19 pandemic spread across the globe, DarwinAI announced the open-source availability of COVID-Net — a deep neural network that examines chest x-rays and allows medical professionals to make  more rapid and accurate diagnoses of COVID-19 cases. It was created using a dataset of 16,756 chest x-ray images from 13,635 patient cases and rolled out to the open-source community in less than a week.

Beyond these two examples, the massive knowledge advantages of AI techniques will be used in the future to augment the highly skilled professionals in the industry, to transform healthcare, and improve outcomes for patients around the world. 

AI in our rainforests

Nestled in the treetops of several rainforests around the world are acoustic monitoring systems, called Guardians, that operate across unimaginably vast distances and act as the ears of the rainforest.

Created by Rainforest Connection — a startup with the goal of stopping illegal deforestation and poaching in real-time — these pieces of Arm-powered technology are made from upcycled cellphone technology and include solar panels, microphones, Arm Cortex-A CPUs, and Mali GPUs. Acoustic data from these devices is run in AI/ML models both in the cloud and on-device, meaning they are energy efficient, reliable, adaptable, and able to alert local authorities of destructive activities or emergencies in real-time.

What’s next for AI?

Looking to the future of innovation like the ones mentioned in this article, we are seeing demand for AI on devices increasing massively. In preparation for the next wave of AI devices, we’ve enhanced our ML support on all of Arm’s processors, including CPUs and GPUs, to deliver significant improvements in ML performance and efficiency compared to previous generations.

Our semiconductor partners are after scalability and flexibility to develop the purpose-built systems they need to run ML in their markets. So that’s what we do — because different computing problems need different mixes of computing components, and we see very clearly that one size doesn’t fit all.

In other systems, partners are increasingly combining many large and small CPUs with GPUs, neural processing units (NPU), and their own IP to generate the right computing solutions for them.

The pervasive growth in Arm-based intelligent endpoints has been staggering as demand for intelligence grows across the wide range of markets Arm’s partners serve, from tiny sensors up to the world’s largest supercomputers.

I really believe we’re just scratching the surface of what’s possible. ML is everywhere in endpoint devices, opening a future of possibilities.