Quote: Original post by Kylotan
But let's not forget that many do not need a fine dining table, only a flat surface they can place a plate upon. :)
If all you want is an approximation to a solution (perhaps so you can get a feel for the rough properties of the solution), then there are simpler methods than ANNs to get them. If you are looking for a solution to a function approximation problem, then an ANN is almost never the best solution. It's just one that works approximately in many situations. In that, yes, it's like tossing a piece of chipboard over two A-frame work horses to make a multi-purpose flat surface. It might be approximately flat, but you probably wouldn't want to serve up a dinner party at it or use it for a game of pool.
Quote: Neural nets and genetic algorithms are certainly overhyped, but there is definitely something to be said for simple, well-known approaches over more accurate solutions that are more complex to understand and implement.
This is one of my personal irks... that most people assume that alternative methods are more complex to understand and implement. This is certainly not the case. Spline methods, for example, are far easier to comprehend and just as easy to implement as an ANN. What scares people off is that many (if not most) of these other methods are not written down in 'lay person books' and they have to hunt around for them in more technical resources. Most people therefore assume that they're hard.
Of course, most of this can be tracked back to societies general aversion to mathematics... but that's a story for another day! ;)
Cheers,
Timkin
