With AI having a real moment in early 2023, I want to take the time to step back and reflect on how I see AI today in IoT deployment and how I see AI tomorrow.
State of AI on the Edge today
Outside of the largest companies, like Apple, most AI on the edge that I see currently is of the unsupervised learning variety. Unsupervised AI is using machine learning techniques on unlabeled data and training an algorithm to learn and identify structures or patterns within the data.
A common problem addressed through unsupervised AI is anomaly detection, which can be very useful on the edge for tasks like detecting potential maintenance concerns prior to a machine breaking down or vandalism detection of a remotely deployed system.
A key benefit of unsupervised learning in the edge context is that you don’t need to transmit the dataset back from the device to leverage this technique. If your edge computer is powerful enough, you can even have it learn about its environment over time, which can create accurate and individualized algorithms on a per edge device or deployment basis.
A lot of the hype we are seeing right now is around supervised learning. So why don’t we see more of that in IoT? It is largely because the costs of transmitting, storing, categorizing, and training data makes the creation of a supervised learning model extremely difficult in the IoT space at this time. Additionally, each device may be running in a different environment, so depending on the actual functions of the device, it’s possible that the data coming from each device may have limited applicability to other devices.
What’s ahead for AI on the Edge?
Unsupervised learning techniques will remain the bread and butter of edge devices for the coming years. As they continue to get more powerful and efficient, and as AI techniques continue to advance, the quality of the results we can obtain from unsupervised learning on the edge is only going to improve.
I do think we will see an increase in the usage of supervised models on the edge and also in the cloud, running across the fleet, as the cost to actually transmit, ingest, and store the data continues to decline, as that is really the limiting factor for most IoT firms at this time.
How can Spinnaker Design help with my AI on the Edge project?
We have the expertise in IoT, AI and Cloud engineering needed to deliver from idea and proof of concept to launch and beyond.
Feel free to comment below or reach out to me on LinkedIn.