computing elements, the neurons, which are connected to each other. When the number of neurons reach into the thousands and the number of interconnections between them reach a hundred, as they did in the Martian mosquito, significant work can be accomplished.
   The technique for training a neural net is to present data to its input sensors. This allows it to process the data in all neurons at the same time by a very simple, hard-coded algorithm. Then the neural net compares its results to the way we tell we would characterize the input. It then adjusts the importance of each neuron to more closely model the expected result.
Truman:
   How long does this training last?
George:
   The Martian mosquito learned our alphabet in one hour. It took several days to distinguish the various soils the planetary scientists had us train it for. Training it to recognize the byproducts of organic life in soil lasted the better part of two months.

 

   This is all out-of-order but that's been the way I've been ever since the robbery. What I mean is that neural networks are modeled after human physiology. Each neuron processes its input independently and in parallel. So neural nets are inherently fast.
Truman:
   Thank you, George.
Dean:
   Mr. Phillips wants to see the Professor's laboratory. George, would you go get it ready for visitors tomorrow?
Exit George
Dean:
   George was Professor Machter's most promising grad assistant five years ago, before George was robbed and shot. At first his recovery was very slow, but the Professor insisted that George return to the lab. About six months ago, he suddenly regained some direction. Even so, you heard his disjointed conversation. His analytical reasoning is fine, but he seems to wander in his thoughts. I don't know what we'll do with him now that his mentor is gone.