What’s been really fun has been working with clients on the development of Internet of Things (IoT) and Artificial Intelligence (AIoT) of Things and discovering the art of the possible. Combining what is traditionally and on-premise, hard wired, material product with a flow of data that provides actionability via data streaming in real-time.
So far predictive maintenance of machines and controller based initiatives to computer vision and real-time object detection have given us the insights to understand the implications of having an EcoSystem, and allows Software Products to interact with hardware products.
One of the fascinating systems is ARM’s ML on the edge (or Edge-ML), combing their processors and other silicon with software libraries optimized to deliver known Machine Learning libraries such as PyTorch, Android NNAPI, Caffee, TensorFlow, etc. This is a serious inflection meaning that a device such as phone or a camera, even a doorbell, would not need to ingest an observation and then send that observation over the internet to an awaiting ML service application to make the prediction on beefy GPU hardware then send it back to the device. But rather the device is able to make the prediction there and then based on the observations and deliver instant output.
ARM likes to use the example of a real-time camera during deep sea diving. Or perhaps another example of instant translation of languages such as that of the translation device seen on episodes of Star Trek.
These ARM architectures in edge-ML processors are being built to be as efficient as possible and very scalable. They operate at near 5 trillion operations per second (TOPs) consuming very little energy.
The future continues to be very bright for machine learning. And obviously there’s an immediate connection with developer operations and analytics. These models need to be deployed, maintained, and analyzed, then incrementally improved. We usually reiterate this is definitely a team approach with loads of new skills to be learned and mastered especially when a traditional in-house development team is being charged with such a new project initiative in AI, AIoT, or IoT.
