What it is interesting is how I got to the numbers. It was an experiment.

I saw in TV an experiment where a hundred people was asked to guess the number of balls in a pot. The results were very spreaded, but the arithmetic average was surprisingly exact.

I tried with the arithmetic average of the results of the other contestants, not having into account the two more spreaded in each category.

The pity is that there were a low number of answers, but even with that, it was close.

Very interesting experiment, indeed.

I’m asking miself it this also works with neural networks, as they work a bit like the brain.

Imagine 100 neural networks, trained with different inputs. May be that the average of results will be much better than only one net, trained with all the inputs and with 100 times the neural cells?

Maybe I try it some day, if I have time.

]]>