NeuroSolutions 6.0 will soon be released to the general public. This new version includes many new features based on your feedback that we are excited to be adding to the software. A summary of these improvements is available below.
As a current NeuroSolutions user, we would like to invite you to try these features before they become available to the general public. We have been using this version for several weeks and are confident it is now ready for general usage.
Since everyone's computers, data, and NeuroSolutions breadboards will have their own unique characteristics, this Alpha will also give us a chance to make sure the new version works well on your
particular model. During this time, we will be paying special attention to making sure problems you encounter with the transition to version 6.0 are addressed as quickly as possible.
NeuroSolutions 6.0 Features
- Mutli-Core/Processor Support - Take advantage of today's high end computing with support for multi-core/processors!
- New Neural Network Topology: Probalistic Neural Network (PNN) - Probabilistic Neural Networks (PNN) are modifications of RBFs that seek to estimate the probability density function by placing a Gaussian function at each input exemplar and multiplying the output of the Gaussian by the binary desired response. General Regression Neural Networks (GRNN) are the same as PNNs except the desired output is continuous. These networks are much faster to train than MLPs for small to medium data sets and often produce better results. PNNs and GRNNs have been included in NeuroSolutions for some time, but their use was somewhat restricted since the weights were set within the NeuralBuilder instead of dynamically within NeuroSolutions. Now there is also the added benefit of having the Gaussian widths be adapted using genetic optimization.
- New Neural Network Topology: Regression Support Vector Machine - Support Vector Machines (SVMs) have now been enhanced to support continuous variables as outputs, allowing them to be used for function approximation problems.
- New Input Optimization: Greedy Search - The Greedy Search algorithm incrementally adds inputs one at a time and keeps those inputs that improve the performance of the network (as measured by the cross validation error). It iteratively performs this process until the addition of any remaining input no longer improves the performance.
- New Input Optimization: Back-Elimination - The Back Elimination algorithm incrementally subtracts inputs one at a time and removes those inputs that hurt the performance of the network (as measured by the cross validation error). It iteratively performs this process until the removal of any input decreases the performance.
Custom Solution Wizard
- Visual Studio 2008 Support - DLL's can now be compiled with Visual C++ 2008.
- Office 2007 Project Shells - Added Project Shells for Excel 2007 and Access 2007.
- Visual Studio 2008 Project Shells - Added Project Shells for Visual C++ 2008 and Visual Basic 2008.
Don't Fret - There's More!
- More Features Coming Soon - This is definitely not the end of the feature list for NeuroSolutions 6.0. We have several more great, highly requested features coming soon in both NeuroSolutions and NeuroSolutions for Excel.
NeuroSolutions 6.0 doesn't end there though - Have a feature that you really want to see in NeuroSolutions or NeuroSolutions for Excel? Take our quick survey to help select new features for version 6.0!
To download and try out NeuroSolutions 6.0, please log into the Licensed User Center by clicking on the icon below. Please be sure to email feedback to support@neurosolutions.com regarding any problems you encounter while using the beta version of NeuroSolutions.
How does Greedy Search work? Greedy Search starts off creating N single input networks for each input your problem contains. It will then "Lock in" the input that produced the lowest cross validation error and begin creating a series of 2-input networks which include the locked in input from the first series. This process will continue to "lock in" the network with the lowest cross validation error until it does not improve the cross validation error any longer.