Re:Reservoir computing using dynamic memristors for temporal...
by captaindiego on Saturday January 06, 2018 @06:50PM
(#16088179)
I think one of the potential advantages is energy consumption of this compared to running deep learning on a conventional CPU/GPU would be dramatically less.
If the memristor chips could be mass-produced, this could lead to things like basic voice commands and face/image recognition being added very cheaply and being added to cheap embedded/IoT stuff (for better or worse).
That being said, I feel like they're over-selling the capabilities of reservoir computing. Yeah, [you can stack them](https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2016...) and get pretty high accuracy on a few tasks, but it's still not competitive with traditional Deep Learning.