Then this is the right place for your contribution!
Most often the quality and quantity of the data acquisition for a project drives its success of failure, I would easily estimate this to 95 % effect on project success.
Or, to put it the other way round: If you have sound and sufficient data for a project then you will succeed easily using RPROP or may be also a hand tuned Backprop with Momentum. That's just my experience, certainly.
Which training algorithm(s) would be of most interest for you? And could you provide some thoughts about why you think that additional training algorithms could be of significant help for users?
Many thanks for your feedback!
thank you for your answer, I don't have much experience with neural networks, I used a few years back Scott Fahlman's Cascade Correlation Neural Network with quickprop and it seemed very good and very fast to converge to a good solution. Also, it requires only a few parameters and grows automatically: each layer has a single neuron trained on a part of the training set.
If I may, where could I find some examples on how to use the C# Wrapper of MemBrain?
Thank you for your time and exhaustive answer.
MemBrain also provides Cascade Correlation (using RPROP during the training phases). You can also find a description on this learning architecture in the MemBrain help (press F1, navigate to the 'Search' tab and and search for 'cascade').neuromante wrote:I used a few years back Scott Fahlman's Cascade Correlation Neural Network with quickprop and it seemed very good and very fast to converge to a good solution.
It sometimes works quite nicely, however, I never found it performing better than a 'hand created' MLP net which is trained just using RPROP. Nevertheless, this may also depend on the used data. If you have a data set at hand which you want to examine then feel free to post it here or send it via EMail. I could also give it a try and compare the results between just RPROP and Cascade Correlation with RPROP.
I like RPROP that much because you really don't have to care about any parameters. It easily adjusts automatically to the used data and net and - in case the data is OK and meaningful - leads to quick and good results.
Have you checked out the C# Wrapper download package from the homepage already? It also includes an example C# project which demonstrates the use of the wrapper classes. Do you have any questions beyond that? Then please feel free to post them here in the forum.TJetter wrote:If I may, where could I find some examples on how to use the C# Wrapper of MemBrain?