Digital neural networks vs. analog resistance meshes
Ever since I first saw an analog-digital hybrid computer in EE lab (1960's), I've been intrigued by using the programmable POTS to solve problems such as the traveling-salesman shortest route. In the digital world this path-following becomes very expensive as the number of nodes increase. In the analog world a preset mesh would calculate the route instantaneously (well, almost.)
It seems that these proposed 8-bit network nodes could be programmed to act like POTS with a huge number of interconnections. I wonder if this could lead towards DNNs taking advantage of some of the analog circuitry with sufficient accuracy.