NeuroGenetic Optimizer 3.0

The NeuroGenetic Optimizer automates neural network design and training by searching through combinations of input variables, neural model architectures and also their internal structures to evolve committees of fully trained high performance robust models that predict what you seek. It is particularly good at time-series modeling.

Easily Create High Accuracy and Robust Neural Networks

Back in the late 1980s and early 1990s, neural networks were having their debut in common applications, but many development tools were complex with many different architectures and internal structures. This made them difficult to use because you really had to be an expert to craft a particular neural net to predict something. Compounding the challenge, it was not clear which inputs to use in the models. The tendency was to use a lot of inputs, or maybe start with a few, but adding inputs sometimes made the results worse because the added input brought more noise than information content. Something was needed to tackle this challenge. As we sat back and thought about it, we realized this is a "combinatorial search" problem, something Genetic Algorithms are good at, and so we created the "NeuroGenetic Optmiizer" (NGO).

A Special Breed of Model Development Systems

The NGO ushered in a new breed of neural network development systems and as such, it is the world’s first general purpose, robust, practical tool to naturally genetically engineer neural networks. The NGO emerged not only from the technical challenge of designing models, but from our consulting services’ need to easily and efficiently discover the best data elements and neural network architectures quickly to build effective neural network applications. In our systems development process, we found we were spending too much time attempting to find the best networks, significant input variables and performing other neural network development efforts manually. It was clear that an effective automation tool was needed to off-load these hours of effort onto computers and hence the NGO was born.

How The NGO Attains High Performance and Robustness

During the search process the NGO splits data into 3 data sets;

• Modeling: used for performing the regression
• Optimization: used for evaluating alternative models and determine when to stop training a model
• Selection: for selecting which models to retain and bring to you in an "ensemble" or committee.

These data sets are extracted sequentially, randomly or random with similarity checking to provide appropriate sampling for model construction. Using regression with "early stopping" (an excellent technique) and double validation and looking for consistency across all three data sets creates models of outstanding robustness.

Customizable

The NGO enables you to add-in specialized data pre-processing using VB.NET or C# using Microsoft Visual Studio Express (available free from Microsoft). Create your own customizations. The possiblities are literally endless.

Use Your Models in Excel

The NGO's models can be used in Microsoft Excel. Just ask us for the "mesh.xls" and we'll email it to you.

NGO Walk-Thru Video

Using the NGO to Find Causes of Problems

Clustering Using the NGO's Self-Organizing Maps

The NGO has the following technical requirements:

• Works on Vista, Windows 7, Windows 8/8.1 and Windows 10
• 4 GB RAM minimum, 8-16 GB is typical
• Minimal hard disk space to install
• Uses .NET 4.5
• Internet connection required to activate and to easily move your license from one machine to another