2020 BRN Discussion, page-261

From the manuscript - a long read !

Training at the edge is clearly a challenge because most devices at the edge of a computing network are typically low power and low performance, which is intentional to make them light weight and enable mobility (e.g., cell phones). Training at the edge would imply placement of training hardware into hand-held or wearable devices, which is challenging in terms of both area and power. The rise of GPUs will be Wave 1 of the three waves of reinventing the computer. We expect ASICs to displace GPUs in the Wave 2 although we do not expect Wave 2 to produce efficient solutions for training at the edge. Alongside these developments, neuromorphic computing would be mature enough to be commercially deployed. Wave 3 will be the deployment of neuromorphic hardware, architectures and algorithms, which will be larger in deployment volume than the preceding two waves, mostly due to the increased demand for compute with time, and also because neuromorphic solutions are expected to address four relevant spaces: inference at the edge, training at the edge, inference at the data centers, and training at the data centers.

Interesting the manuscript takes account (references) of March 2019 BRN Update.

A lot has happened since then as Akida at the Edge gets closer to reality and we have a clearer idea what it can do at the Edge - Unsupervised learning DNNs and run lots of algorithms

The bit that intrigued me was the pace (waves) they expect computing to improve.

From my take on it all Akida is the "very early" delivery of Wave 3 thereby jumping waves 1-2- impressive