Thank you for Subscribing to CIO Applications Weekly Brief
Thank you for Subscribing to CIO Applications Weekly Brief
The advent of IoT brought forth a revolution in embedded systems connected to the cloud. In that perspective, it was noted that while cloud computing offers unprecedented economies of scale, it is falling short of o making the most out of our embedded systems. The volume of data generated by embedded devices makes real-time cloud processing difficult and expensive due to the high bandwidth, cost, and pervasive availability. Not to mention the data will also be governed by security, privacy, and regulatory constraints. Embedded ML, one of the most talked-about tech advancements for smart devices, provides a solution by allowing complex computations to take place on the embedded system instead of being sent back and forth to a cloud server. This approach enables developers and product designers to build machine learning models that can infer actionable data, locally. Such distributed edge models can work better in many use-cases as they ensure better privacy, tighter security and faster speed. In addition to drastically reducing power consumption, this approach requires little to no network connectivity.
Edge Impulse has established itself as the leading development platform for machine learning on edge devices, offering an end-to-end platform for developers to create, train, deploy, and manage ML models efficiently on nearly any embedded device imaginable. With their platform, developers can build ML models in real-time tapping smart device’s sensors and MCUs, from the accelerometer, microphone, or camera to collect data and train machine learning algorithms in a continuum. We recently interviewed Zach Shelby, Co-founder, and CEO at Edge Impulse, who shed light on how they enable the ultimate development experience for machine learning on embedded devices and why their model can help enterprise companies scale and re-imagine what was never possible before.
Could you provide us a brief overview of Edge Impulse?
In the early 2000s, I co-founded one of the world’s first IoT startups. During my time at the startup, I worked on a new technology area that involved connecting edge devices with the internet securely. Later, the company was acquired by Arm in 2003, and I became a VP at the firm. During this period, we tried to build solutions to solve industrial problems using connected edge compute. We understood that companies fail to leverage the vast amounts of data generated by IoT and embedded systems. My partner Jan Jongboom and I realized that we could use edge computing capabilities to leverage the data generated by embedded computers. Later, in 2019, we founded Edge Impulse, with a mission to enable developers to create the next generation of intelligent devices using Embedded ML.
The computing power of embedded processors is growing at exponential levels. Owing to advancements in technology, embedded processors have become relatively cheap and battery efficient. Our aha moment came when we realized that we could utilize these embedded systems' computing power to harness the power of data. We started to apply ML models in the space of edge computing. ML makes data more valuable as it turns data into actionable insights. We can use ML in industries that traditionally require us to perform tasks such as predictive maintenance, asset tracking, occupancy detection, and health and safety measurements.
Once we realized that we could harness the computing power and data of billions of devices by coupling them with purposeful machine learning models, we knew that we will change embedded product design for good
Sensors today work at a very wide bandwidth due to the advancements in microcontrollers, CPUs and connected mechanical hardware. For that reason, one can collect massive data out of the sensors at higher bandwidths. However, companies struggle to move this data to the cloud with the existing IoT technology. Such companies approach us for the know-how to make the most out of the information they collect. We help them in organizing the data they collected by building the algorithms. We give the engineering team valuable data sets and enable them to work on industry problems.
Giving access to signal processing and machine learning algorithms is yet another challenge we solve for our clients. After building the algorithm, sharing it across the engineering team is one of the biggest problems. For that, we help our clients in breaking down the siloed R and D team within an organization. Typically, ML in the cloud space is an extremely power-consuming technology. We are dealing with huge data sets and big ML models, and they are not optimized to work efficiently. Running them on edge devices is an uphill task. All of our clients experience a hard time to get access to the tools that let them run the models they build. To bypass this, we give them pre-optimized tools. We optimize the signal processing for edge compute before deploying the models. With that, we also solve deploying the interference library that runs on devices including low power microcontrollers such as Arduino devices, and high-power consuming devices such as embedded Linux devices, and Raspberry PI class devices.
Could you narrate an instance when a client approached you with a unique challenge and Edge Impulse helped them solve it using its ML capabilities?
After doing a market study, we found a high demand for Edge ML in areas including predictive maintenance, condition monitoring, and smart city infrastructure. Recently, we worked with a customer that provided installation services for electric power poles and power lines. The client uses an intelligent sensor and insulators, which monitors the power line's electrical and non-electrical parameters in real-time. Typically, one can monitor only a few parameters because of power consumption constraints. By applying an embedded machine, the client successfully observes complex phenomena like lightning strikes, potential fire, hitting lines, possible damages to the poles, and more. They can send an alarm to the grid maintainer to avoid damages. It is worth mentioning that they can do it continuously for many years as we ensured to provide a battery life of ten years. Giving developers the tools to train, reproduce and enrich their model is one of the critical functions of our platform, followed by the ability to scale, manage, and our “no-code deploy” technology that offers a dependable and continuous management of your target hardware and ML models (MLOps).
What makes Edge Impulse unique in the ML and AI space?
The software development industry is replete with many machine learning tools targeted at data scientists primarily. However, when it comes to working on physical solutions for areas such as asset tracking, and health, it is the developer who gets the job done. What makes us different is that we focus primarily on the software developer who practically works on the product. However, we ensure that they get help from data scientists if required. If they need help with a new algorithm, we have a way to interact with data scientists. As we focus on the developer, we are good at solving problems for them. We provide a user experience that enables a developer who uses their data to solve the problem on a real device. The specialized data science tools help a different group of people, not the developers. For that reason, we adopt a developer community focused go to market strategy. As we offer an open developer platform, anybody can create an account on edge impulse, deploy the solution and start building machine learning models free of cost. At present, we have a vast community of 6,000 developers who work on around 10,000 projects on Edge Impulse. Over a thousand enterprises use our product, and they have created more than 10 million data points in our datasets.