Neural network reproduces what it has learned. If anything resembles something it has learned, the response will be similar ( thank you continuous functions).
I remember I made a neural network to read my handwriting, then any handwriting. It worked really well. No wonder they use it in the postal services… achieving a better result than human eye.
From that outstanding result, I tried to apply it everywhere. After all, isn’t our a brain a neural network ?
As I was involved in the Euromanager contest, where the game was to manage a virtual firm, I noticed that all the finalists, the year I participated, were engineering students ( at the beginning it was half engineering students, half business students). Everyone was using computer power in order to break the code… We tried to use neural network from an extensive data history in order to optimize the algorithm. We used the balance sheet, the cash flow table, the profit and loss as inputs, taking into account history. I later tried it on real cases.
Needless to say, I had very poor results, even if we arrived 4 out of 132, won bottles of champaigs and some very nice presents. It also gave me Banking job opportunities.
I now believe that neural networks, optimization algorithms, are a statement of failure.
It is refusing to understand the principle, the underlying function, and going the shortest way to the answer.
Unfortunately, the shortest way sometimes leads nowhere. Functions might not be C°, or anything.
Why is our brain different ?
I don’t know. And if I knew I’d be a Nobel prize and use my million to buy some fancy ties. (I can’t even afford an Airbus or a Boeing at this price).
| 
| 
| 
| 
| 
| 
Vinhly – Fresh vinhlys on Mondays and Thursdays