Bill,
When I got my B.S., 16 years ago, Artificial Intelligence was considered the poor step child of Computer Science. It was considered just interesting theory, with little practical application. I might even say there was a bit of a stigma attached to AI, for showing so much promise in the 70's and consuming so much research time and money without producing much in the way of tangible results. It's only more recently, as more computing horsepower has become available, that the applications are becoming reality. A common Pentium PC has more computing power than the mainframes of that era.
I'm learning this NN stuff as I go. I concur with your description of how backprop nets work, except the 'bias weights' aren't something I remember seeing before.
I think the technology should work, I'm still cautious about declaring that it does work, only time will verify the results. I was becoming somewhat uncomfortable with the black-box nature of NeuroStock even before these other issues were fully disclosed. I find that I prefer the approach Neuroshell uses, but it is to a large degree based on conventional TA indicators. I would not advise jumping into it unless one has a working knowledge of conventional TA. Building systems in NeuroShell has caused my to pull out all my TA books and start studying them again.
I am still hoping that I will be able to use NeuroShell and NeuroStock as complimentary systems, confirming each others' signals.
CL
|
