Search results

Nonsense! Properly constructed neural networks do not "go haywire". Overfitting is a beginner's mistake. If training is controlled, overfitting will not occur.

Why rule induction instead of something else? Why rules discovered by genetic algorithm, specifically?

I don't understand why you believe that's the case: There are certainly ways to constrain neural network training to avoid this problem.

This is easily remedied through any number of means: early-stopping, limiting the number of hidden neurons, etc.

Could you clarify this? Is it the fact that a system is based on neural networks or merely that is automated that leads you to conclude that it will not work?

"Near impossible" or "impossible"? If "impossible", then this wouldn't this conclusion apply to any attempt to pick a turning point, not just neural networks?

Fuzzy logic is an uncertainty management system which is premised on the idea that items may partially be members of some sets (fuzzy sets). While the fuzzy calculus can be applied to many analytical procedures (like fuzzy clustering or fuzzy regression), in practice, fuzzy logic most often is...

Could you explain further what that means?

While neural networks can be trained by genetic algorithms, they are much more often trained by some other, more direct procedure (gradient descent, for instance). The important thing is to understand that genetic algorithms, in themselves, do not "learn" anything- they are optimizers of other...

None of this is relevant. For a claim that "neural networks can never be used productively for trading" to be proven beyond doubt would require an impossibly exhaustive experimentation. The most one could seriously claim is something like "[assertion]: no examples of neural networks aiding...

If I may ask: What do you do for a living?

Wait: BioComp is a software tool, not worked-out solution. That's like saying "Using saws to make birdhouses doesn't work. Craftsman is an example." Having a tool and using it to solve a problem are two completely different things. None of this is to say that advocates of neural networks...

??? Could you explain this? Genetic algorithms and neural networks perform different tasks. Genetic algorithms are optimizers, whereas (most) artificial neural networks are classifiers or regressions.

While you are correct, this should never happen is models are being constructed appropriately. All the test data (by whatever method) provides is an unbiased estimate of model error. Simplistically, assume that some test indicates that our model's tested performance is within specified...

Solutions to this problem need not be arbitrary. This subject has been well-studied and documented in the literature. Other resampling methods include k-fold cross validation and bootstrapping. I recommend Weiss and Kulikowski's Computer Systems That Learn. -Will Data Mining in MATLAB