Edge Impulse Imagine 2022 Proposes Innovative Approaches to Further Empower Edge AI Developers

Subscribe To Download This Insight

4Q 2022 | IN-6729

Edge Impulse Imagine 2022 gathered some of the big players in the edge Machine Learning (ML) and TinyML markets, and showcased myriad solutions, partnerships, and technological innovations. The event also highlighted some of the challenges ahead, not least the need for a huge increase in the number of ML developers at work today, as well as the number of ML workloads.

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.


Toward a Million Developers in TinyML


Imagine 2022, organized by Edge Impulse, a leading development platform for Machine Learning (ML) at the edge, was back in town in September, and for 3 days, some of the leading players in edge ML and TinyML gathered at the Computer History Museum in Mountain View, California, to showcase their current solutions, as well as their latest collaborations and partnerships. An event that featured edge ML demonstrations in the sectors of supply chain logistics, smart home living, smart healthcare, and computer vision most prominently, there were also multiple talks and presentations, various workshops, and even competition prizes from Seeed Studio, Arduino, and Sony.

Perhaps most remarkably, however, was the vision Edge Impulse presented at the event. Starting from the claim that the ML market in the enterprise sector is valued at US$250 billion, combined with a forecast from ABI Research that, by 2030, the TinyML market will be worth US$1 billion, Edge Impulse envisions a future in which there should be 1 million ML developers for 10 million ML workloads, from the 100,000 workloads and 50,000 developers of today (more about the plan to achieve this is below).

Edge AI Everywhere


If nothing else, Imagine 2022 was full of announcements of some of the latest technological developments in the TinyML sector. In the keynote talk, Edge Impulse announced the Health Reference Design tool (a clinical data collection and on-device ML solution), the Machine Reference Design feature (an end-to-end edge ML application), and the Assisted Labeling algorithm, in addition to various partnerships—in particular, the launch of the “White Label” program, with Arduino as the first such partner, making Edge Impulse Studio available to White Label partners as a fully customizable tool, thus fostering deeper integration, and a partnership with Balluff (maker of sensors) and Ready Robotics to apply small ML models in anomaly detection directly on robots.

And from the rest of the participants, there was the presentation of the ML-powered, Conservation X Nature camera, with ML available on all of its sensors and with a 10-year battery, making it apt for the collection of data in the wild; the sensors of NOWATCH, a Dutch startup, to measure cortisol levels, which can then be combined with ML models to work out stress levels; the non-invasive ML approach of Know Labs to monitor blood glucose; the configurable, programmable analog chip by Polyn Technology; the Optra platform from Lexmark, meant to monitor the functions of a printer from different perspectives (e.g., the life of the printer’s fuser, the timing involved in paper picking, etc.); Synaptics’ presentation of the Katana System-on-Chip (SoC), an ultralow power edge ML chip; BrainChip’s demonstration on how to leverage its neuromorphic processor, Akida, with Edge Impulse Studio; and many others.

When Tiny Pays Off


As mentioned, a remarkable moment from Imagine 2022 took place right at the beginning, during the keynote, when the Chief Executive Officer (CEO) and co-founder of Edge Impulse, Zach Shelby, played a blinder and expounded on the vision to reach 1 million ML developers and 10 million ML workloads in the TinyML space. This is a target to be reached on the back of the current 50,000 developers and over 100,000 ML workloads. However, it is unclear if these numbers refer to the Edge Impulse ecosystem alone or to the TinyML market as a whole—and in line, possibly, with ABI Research’s own estimate that the TinyML market will be valued at US$1 billion by 2030. This forecast is not based on what ML workloads and developers are needed to make this a reality, and the focus on such round numbers as 1 and 10 million may not be substantial in and of themselves, but simply aspirational because of the scale.

Be that as it may, the plan Zach Shelby laid out at the event is an interesting one, and a threefold one at that. First, vendors and developers should adopt a sharing and cloning approach to workloads, and even to algorithms, an approach not too dissimilar to the GitHub model in the development of Open Source everything to promote innovation, transparency, and collaboration across the board. Second, developers ought to be able to employ a “bring your own model” approach to applying and sharing ML models, and in the case of Edge Impulse, this would mean that any training pipeline could be hosted in its Studio ecosystem, an approach that can be based on a Python Software Development Kit (SDK) for TensorFlow and PyTorch. And third, the wider adoption of Edge Impulse’s University Program, which already boasts 300 members, including renowned universities like Harvard, as well as some important market players like Arduino and Arm, is meant to increase the number of edge ML professionals.

It is uncertain what success this plan will have toward realizing Edge Impulse’s vision. Nevertheless, Edge Impulse is certainly well placed to play a central role in this endeavor. It has developed developer friendly tools, including low-to-zero code tools. It can support a wide range of applications, from computer vision to audio and ambient sensing, encompassing the full spectrum of edge ML applications, while maintaining active community engagement and industrial partnerships, some of which have been mentioned here. Moreover, what is clear from events like Imagine 2022 is that the possibilities for edge ML and TinyML are enormous. Rather than applying ML to big problems that may not be amenable to ML  anyway—think of the work of radiographers, which contrary to some wild predictions, is a pretty safe job for humans for some time to come—applying ML to the edge to try to solve small problems, most often involving environmental sensors of various types, holds greater prospects. Consider any kind of sensorial data from the environment that can be attended to and there is probably an ML model that can be applied to that data—and it is here that size doesn’t matter (or not as much).