Introduction

Neural networks is a very sophisticated modeling technique capable of modeling extremely complex functions. Neural networks creates these models by interconnecting a number of relatively simple neurons. The resulting network transcends the simplicity of the neurons to solve complex solutions.

An advantage of neural networks is its raw strength. It can model very complex data. Its models can work in classification, regression, association, clustering, and time-series tasks.

A disadvanage of neural networks is that its power is based on the structure of its neurons. Guessing the proper structure is either left to an expert to artfully weave, to trail-and-error, or most commonly a bit of both.

---

Features

EWA's Neural Network Engine supports any network that can be described as a directed acyclic graph (any connection of neurons that when connected with arcs feeding forward through the network will not create any cycles), with the exception that cycles are permitted in time-series problems.

• Implemented in 100% Java, with performance similar to C.
• Unlimited Problem Size (Problem does not have to fit in memory)
• Uses EWA's Standard Data Manipulation Package
  • Complete Data Selection Features
  • Merge/Join of Multiple Tables
  • Full Data I/O Capabilities
    • ODBC, XML and ASCII Data Files
  • More about the Data Manipulation Package...
• Uses EWA's Standard Data Preparation Package
  • Supports Separate Learning/Testing Datasets
  • Supports Single Learning/Testing Dataset
    • Dedicated Test Set
    • Verification Folding
  • Supports Data Subsampling
    • Sampling from dataset's start, end, evenly, or randomly
  • More about the Data Preparation Package...
• Configuration Options
  • Learning Rate
  • Learning Relaxation Rate and Period
  • Learning Momentum
  • Testing Periods
  • Min and Max Stopping Periods
  • Stopping Error Rate
• Results
  • Neural Network (including neuron weights)
  • Confusion Matrix
• Neural Network Data Sets and Resulting Models can be persisted to XML

The Neural Networks Engine includes a comprehensive GUI controls, manipulating the network at the network, layer, neuron, and neural link levels. The following screen captures shows a network being built using the GUI:

---

Performance

In a third-party single threaded comparison, EWA's Neural Network Engine had performance much faster than other Java implementations. All implementations achieved similar error rates.

	EWA Systems (Java)	WEKA (Java)
Sleep Data Set	12m with 25MB	55m45s in 123MB

---

Competitors

• Neuralware

---

Additional Reading

The Basics

• Carling, A., Introducing Neural Networks. Wilmslow, UK: Sigma Press, 1992.
• Fausett L., Fundamentals of Neural Networks, Prentice-Hall, 1994.
• Gurney K., An Introduction to Neural Networks, UCL Press, 1997.
• Haykin S., Neural Networks, 2nd Edition, Prentice Hall, 1999.

More Advanced

• Bishop, C., Neural Networks for Pattern Recognition, Oxford: University Press, 1995.
• Patterson, D., Artificial Neural Networks. Singapore: Prentice Hall, 1996.
• Ripley, B.D., Pattern Recognition and Neural Networks. Cambridge University Press, 1996.

---

To Purchase or For More Information, contact our Sales Team.