Neural net

There are 3 input neurons on the left (transistors). You can simulate the signal received by each using the vr on extreme left. Each neuron has a dendrite connecting to the output signal (lamp). Each dendrite can be assigned a weight using one of the 9 vrs. To calculate the weights for an application would be done on a simulator for a given application, and correct resistors manufactured to this specification for the application. This type is called an accelerator since it could operate at orders of magnitude faster than a software implemented neural net. Currently accelerators convert back to digital for each level of the network but this is an expensive operation. The idea is that using transistors or Mosfets at each level may allow several levels to be accomplished in the analogue domain. Therefore you could build lightning fast neural networks, but they would be aplastic and hard coded, without the ability to learn.