Why Edge AI for Wearables?

This is the second of a series of articles exploring the benefits of Edge AI for a variety of applications. 

Wearable technology, or "wearables", refers to electronic devices that are worn as accessories such as glasses or watches, contain processing and computational capabilities, and some form of connectivity^[1]. They also include devices embedded in clothing, implanted in a consumer’s body, or even tattooed onto their skin.

Wearables have experienced an incredibly rapid growth in recent years. Between 2016 and 2022, the number of connected wearable devices worldwide skyrocketed 240% from 325 million to 1.1 billion^[2].

However, a lot of the data and information processing these devices mediate is still done centrally on the cloud, or on a connected smartphone. This leads to high computing power consumptions, increased emissions, decreased battery life, and privacy concerns.

In this article, we discuss how Edge AI can address these issues by making decisions not in the cloud, but instead on the device itself - or on the “edge” of our analog and digital worlds. Edge AI helps wearables' battery life last longer, reduces latency and cloud communication delays, and facilitates a two-way interactive relationship between device and user, where it was previously more one-sided.

What is Edge AI? 

Traditionally, artificial intelligence is often associated with the cloud, a centralized location where data is collected and AI algorithms are developed and run for processing or making decisions based on the data. By contrast, edge AI refers to the application of AI as close to the end user as possible - where the computation is done on the "edge" of the network, close to where the data is located and captured, rather than centrally in a cloud computing facility^[3].

The reach of the internet is global, meaning data can be captured anywhere. This makes it theoretically possible for the "edge" of a network to be any location on the planet. Depending on the data being captured, edge AI might be deployed in a hospital, a 5G cell tower, a retail store, on traffic lights, and on personal devices like phones, laptops, or wearables.

Edge AI is a powerful tool that can bring digital transformation practices that are commonplace in the cloud close to the physical world^[4]. It can be used to improve efficiency, performance, management, and security of personal devices using cloud best practices. 

How does Edge AI Work? 

Edge AI is an exciting development because it allows traditional technologies to run more efficiently, with higher performance, and on less power. It relies on neural networks and inference algorithms to bring AI development outside of the cloud^[5].

Neural networks are often used to deploy the most complex of machine learning algorithms, known as deep learning, and require a lot of data and computational power to be adequately trained^[6]. Furthermore, the data provided to train this network has to cover examples of positive and negative versions of what you're trying to teach your network to decipher between. For example, if you're trying to teach a wearable heart rate monitor such as a FitBit to detect when your heart rate has risen to a dangerous high, you need to feed it data on healthy heart elevation rates as well as dangerous rates, so it learns to distinguish. This training - or deep learning as it is known - often takes place in the cloud or powerful GPUs due to the amount of data needed to train a model.

Once the neural network is sufficiently trained, it can be taught to do inferencing. Inferencing is the act of using a trained neural network to provide insights independently without intervention from a human^[7]. It isn't dissimilar to a student going from training to be a teacher to working in a classroom in the real world. The "graduated" trained models can now answer real-world questions relating to the data they were trained on.

Training the network takes a lot of computational power but inferencing only requires a fraction of it. This means that it should be possible to inference on the edge, with devices that are purposefully designed to offer the right performance, while consuming very little energy so they can be deployed in a wearable device. For example, neural decision processors offer more than 100x efficiency improvement while providing over a 10x increase in throughput compared to typical MCUs, making them ideal for inferencing on the edge.

Why Use Edge AI for Wearables?

A rapidly growing application is the use of edge AI on wearable devices, specifically wearables targeted towards health and fitness. 75% of users agree that wearables help them keep track of their health^[8]. 

Using edge AI on wearables provides a host of benefits compared to relying on traditional cloud-based AI.

• Low power: A self-learning AI can update its edge AI framework without access to the cloud, requiring much less power to operate^[9]. This is because deploying a trained neural network on the edge and allowing it to inference there, rather than in the cloud, means it requires significantly less computational power. This is great for companies and users, since there is no need for expensive, power-hungry hardware, and the device's battery life is extended considerably^[10].
• Less latency, more real-time results: Inferencing locally eliminates delays that could arise with cloud communication, or the possibility of central servers being down. Using a low-power computer and inference accelerator close to the source of data means much faster response time. Cloud-based response time is usually a few seconds. Edge AI takes this down to milliseconds^[11]. 
• Increased reliability: Since edge AI does not rely on communication back and forth from the cloud, it is able to work autonomously despite technical issues that might happen on the network or the cloud
• Stronger security and data protection: EdgeAI applications minimize the amount of information transferred between the device and the cloud. This reduces the attack surface of the AI system, which leads to stronger security and fewer opportunities for data breaches. By reducing data beaches, EdgeAI provides a clear value proposition to wearables applications that handle sensitive data in areas such as smart cities and healthcare.

Edge AI can provide secure conditions for training and maintenance with mixed reality wearables.

Conclusion

Edge AI brings hyper-personalized, ultra-low-power solutions to end users by bypassing the cloud and carrying out all inference locally at the edge. It enables pervasive wearable devices to access the compute resources they need, while ensuring strong security and real-time performance. Moreover, local inference allows for a smoother user/device interaction, for example through voice-activated and touch-free wearables. 

Designing Edge AI wearables with Syntiant

California-based edge AI company Syntiant makes ready-to-use technology that allows users to interface naturally with their wearable devices^[12]. For example, Syntiant empowers users to interact with their devices through a speech interface, or gestures, without having to navigate a menu on a small screen.

Syntiant’s NDPs technology can enable any device with always-on neural processing capabilities, offering unparalleled power and performance. Specifically, Syntiant’s end-to-end solutions bring a new level of machine learning to wearables, which was previously only available at 10x the power and at least twice the die size.

Syntiant technology enables the integration and execution of low-power, performance savvy applications into smartwatches, fitness trackers, health monitoring wristbands and smart textile devices. As a prominent example, Syntiant NDP technology empowers wearable devices OEMs (Original Equipment Manufacturers) to integrate robust speech processing capabilities for effective voice commands. Likewise, it can also enable the integration of advanced motion detection functionalities that can greatly enhance the intelligence capabilities of the wearable device. 

Overall, Syntiant facilitates the implementation of small-size, low-power wearable solutions that provide unique user interfaces. The latter enable solutions that are ergonomic, extremely easy to use, and stand out in the market. Syntiant’s technology is ready to integrate and use within real-life wearables applications.

The first article discussed about reducing AI's Vulnerable Attack Surface with Edge Computing.

The second article talked about Edge AI in wearables.

The third article explored how edge AI is enabling cutting-edge advances in sustainability.

The forth article explained why Edge AI is a win for automotive.

The fifth article analyzed computer Vision on Compute-Constrained Embedded Devices.

The sixth article explained why edge AI is essential for EV Battery Management.

About the sponsor: Syntiant

Syntiant Corp. is a leader in delivering end-to-end deep learning solutions for always-on applications by combining purpose-built silicon with an edge-optimized data platform and training pipeline. The company’s advanced chip solutions merge deep learning with semiconductor design to produce ultra-low-power, high-performance, deep neural network processors for edge AI applications across a wide range of consumer and industrial use cases, from earbuds to automobiles. Syntiant’s Neural Decision Processors™ typically offer more than 100x efficiency improvement, while providing a greater than 10x increase in throughput over current low-power MCU-based solutions, and subsequently, enabling larger networks at significantly lower power.

References

1. Hayes A. The Ins and outs of Wearable Technology [Internet]. Brock T, editor. Investopedia. Investopedia; 2022 [cited 2022Oct6]. Available from: https://www.investopedia.com/terms/w/wearable-technology.asp 

2. Laricchia F. Global Connected Wearable Devices 2016-2022 [Internet]. Statista. 2022 [cited 2022Oct6]. Available from: https://www.statista.com/statistics/487291/global-connected-wearable-devices  

3. Yeung T. What is Edge Ai and How does it Work?  [Internet]. NVIDIA Blog. 2022 [cited 2022Oct6]. Available from: https://blogs.nvidia.com/blog/2022/02/17/what-is-edge-ai/ 

4. Lawton G. What is Edge AI? [Internet]. SearchEnterpriseAI. TechTarget; 2021 [cited 2022Oct6]. Available from: https://www.techtarget.com/searchenterpriseai/definition/edge-AI 

5. Ibid

6. Ceron R. AI Today: Data, Training and Inferencing [Internet]. Servers & Storage. 2020 [cited 2022Oct6]. Available from: https://www.ibm.com/blogs/systems/ai-today-data-training-and-inferencing/ 

7. Yeung T. What is Edge Ai and How does it Work?  [Internet]. NVIDIA Blog. 2022 [cited 2022Oct6]. Available from: https://blogs.nvidia.com/blog/2022/02/17/what-is-edge-ai/ 

8. Samet A. Wearable technology trends that will shape healthcare in 2022 [Internet]. Insider Intelligence. 2022 [cited 2022Oct6]. Available from: https://www.insiderintelligence.com/insights/top-healthcare-wearable-technology-trends/ 

9. Flaherty N. First self-learning edge AI sensor for Wearables - Video [Internet]. EENewsEurope. 2020 [cited 2022Oct6]. Available from: https://www.eenewsanalog.com/en/first-self-learning-edge-ai-sensor-for-wearables-video/ 

10. Woolley A. What is AI Inference at the Edge?: Insights: Steatite [Internet]. Steatite Embedded. 2020 [cited 2022Oct6]. Available from: https://www.steatite-embedded.co.uk/what-is-ai-inference-at-the-edge/