Press the ESC key to leave the presentation at any time.

1
Select a product for more information.
Press the ESC key to leave the presentation at
any time.
2
Help
Navigating This NeuroDimension Presentation
Simply select the topics you would like to view
and then use the arrows to advance through the
information on that topic.Select More Info to
view additional detailed information.When you
are done with a topic, simply press the up arrow
to return.
Select the Up Arrow below to return from this
help screen.
Select the Left and Right Arrowsto move through
a topic.
3
NeuroSolutions is one of the slickest and most
complete packages for neural network simulation
that anyone could wish forand one of the most
flexible as well -Dan Ellis IEEE Spectrum
Click here to return to the ND main menu.
4
NeuroSolutions
The Neural Network Simulation Environment
NeuroSolutions provides an object-oriented
simulation environment for neural network design
and application. It has quickly evolved into the
software tool of choice for both the neural
network beginner and expert alike. This leading
edge software combines a modular, icon-based
network design interface with an implementation
of advanced learning procedures, such as
recurrent backpropagation and backpropagation
through time.The result is a virtually
unconstrained environment for designing neural
networks to solve real-world problems such as
financial forecasting, pattern recognition,
process control, targeting marketing, and many
more.
Select any of the following topics for more
information or use the arrows to step through
them all.
Graphical User Interface
Interactive Probing
Advanced Features
Neural Network Creation
5
Graphical User Interface (GUI)
NeuroSolutions is based on the concept that
neural networks can be broken down into a
fundamental set of neural components. By allowing
the user to arbitrarily interconnect these
components, a virtually infinite number of neural
models can be constructed.
Neural components, such as axons, synapses, and
gradient search engines, are laid out on a
graphical breadboard and connected together to
form a neural network. Input components are used
to inject signals, and probe components are used
to visualize the networks response.
6
Neural Network Creation
Creating neural networks is fast and easy with
NeuroSolutions. Let the NeuralExpert create
and customize a neural network for your type of
application. Or, let the NeuralBuilder build a
neural network topology to your specifications.
Plus, enjoy the flexibility of being able to
create neural networks directly from palettes of
customizable components or modifying existing
designs.
Select either of the following topics for more
information.
NeuralExpert
Application-based neural network designer.
NeuralBuilder
Topology-based neural network creator.
Data Manager
Data management and analysis tool.
7
Interactive Probing
Probing is an important step in the neural
network design process and is therefore, an
integral part of NeuroSolutions. As with the
neural components, the probe components are
inherently modular the way you view the data is
independent of what the data represents.
NeuroSolutions probes provide you with real-time
access to all variables during the simulation,
along with a variety of ways to visualize them.
This represents an unparalleled ability to see
what is going on inside your networks.
8
Advanced Features
Advanced users will want to utilize the software
to the next level. Researchers will invariably
want to integrate their own algorithms into
NeuroSolutions application developers will want
to integrate NeuroSolutions algorithms into their
own and those prototyping large networks within
NeuroSolutions will often want to run them on
faster hardware platforms. NeuroSolutions was
designed to accommodate all of these
requirements.
Select any of the following topics for more
information.
Genetic Optimization
Macros
Sensitivity Analysis
OLE Automation
Dynamic Link Libraries (DLLs)
Code Generation
9
Probing
Matrix Viewer
Neural networks are often criticized as being a
black box technology. With NeuroSolutions
extensive and versatile set of probing tools,
this is no longer the case. Probes provide you
with real-time access to all internal network
variables, such as Inputs/Outputs
Weights Errors Hidden States
Gradients Sensitivities Probing
is an important step in the neural network
design process, therefore we have made it an
integral part of NeuroSolutions. All network data
are reported through a common protocol, and all
NeuroSolutions probes understand this protocol.
This provides you with access to all internal
variables, along with a variety of ways to
visualize them.
Bar Chart
Matrix Editor
State Space
Data Writer
Image Viewer
Scatter Plot
Data Graph
Hinton Diagram
Spectrum Analyzer
10
Probing
Matrix Viewer
Matrix Editor
This probe displays the current data values in
matrix format, providing quantitative information
about the data being probed. It can be used to
obtain the value of any internal network variable.
Similar in function to the Matrix Viewer, this
probe has a very important additional function
it allows you to modify the data being probed.
For example, you may want to modify the weights
or inject a specific pattern to determine how the
network responds.
11
Probing
Data Writer
Scatter Plot
The Data Writer collects and displays data in
matrix format during a simulation. The collected
data may then be written to an ASCII or binary
file, which can be used for further processing or
reporting results.
This probe plots the temporal data of one channel
(PE) against the temporal data of one of the
other channels. Multiple pairs of channels can be
specified. The data from each pair is used as the
X and Y coordinate of a two-dimensional graph.
The points are collected over a number of samples
to produce a scatter plot.
12
Probing
Data Graph
Spectrum Analyzer
This probe can be used as a multi-channel graph,
displaying amplitude versus time. Typical uses
include displaying time-varying network activity
and network learning curves. This is a necessity
when working with temporal problems such as time
series prediction.
The Spectrum Analyzer is used to compute
periodograms from temporal data. Periodograms are
generated by averaging windowed Fast Fourier
Transforms (FFT's) over time.
13
Probing
Bar Chart
State Space
This probe provides a qualitative feel for probed
data in the form of horizontal bars. The length
of each bar represents the magnitude of the
signal at one channel. This is very useful for
static classification problems when making a
comparison between an output and its desired
response, or as a thermometer of the networks
performance when probing the mean squared error.
The State Space probe displays a 3-D
representation of the systems state as it
evolves over time. It does this by plotting the
signal against approximations of its first and
second derivatives. This tool is very useful for
dynamic system analysis.
14
Probing
Image Viewer
Hinton Diagram
The Image Viewer interprets and displays network
data as a bitmap image. The data is normalized
and converted to a matrix of values corresponding
to pixel intensity levels (white corresponds to
one and black corresponds to zero).
The Hinton diagram provides a qualitative display
of the values in a data matrix (normally a weight
matrix). Each value is represented by a square
whose size is associated with the magnitude, and
whose color indicates the sign.
15
NeuralExpert
NeuralExpert
Application-based neural network designer.
Do you have a specific problem that you would
like to solve with neural networks? Let the
NeuralExpert design and customize a neural
network topology around your data and the
solution you would like to find. The NeuralExpert
eliminates the need to know which type of
topology is best for your problem type by
selecting the right design for your application
and customizing it to your needs.
Simply select the type of problem you would like
to solve, and then answer a few questions about
your data and how you would like to process it.
The NeuralExpert will select the appropriate
neural network topology based on the problem type
and customize it for your application-specific
data. The created network will typically be
capable of solving your problem. However, since
it is created in NeuroSolutions, any part of the
neural network topology can be updated or
modified.
16
NeuralExpert
Step 1
Using the NeuralExpert
Problem Type Selection
The NeuralExpert makes it easy to create the
appropriate type of neural network for your
application. Simply select type of problem you
want to solve and provide it with a sample of
your data. The NeuralExpert automatically
tailors its questions to the problem type. There
is even a beginner level that allows you to
skip over advanced options that typically dont
need to be selected. Here is an example of the
steps taken to create a neural network to
classify the sex of crabs from their
characteristics
Select the type of problem you want to solve from
the following types of problems
Classification Function Approximation
Prediction Clustering
17
NeuralExpert
Steps 3-4
Step 2
Tag Input Columns
Input File Selection
Indicate which columns in the input file to use
as inputs for your problem. You can even use
symbolic or categorical data such as Male and
Female in the input columns!
Indicate where your input file is located. The
NeuralExpert will customize the solution for your
specific inputs.
18
NeuralExpert
Steps 6-7
Step 5
Tag Desired Columns
Desired File Selection
Indicate which columns in the desired file to
model. As with the inputs, you use symbolic or
categorical data in any of the columns.
Indicate which data to use as an example of the
data you are trying to model. This can be a
separate file or the same file as the input file.
19
NeuralExpert
Use Your Neural Network
Step 8
Set Network Complexity
Based on your data and selections, the
NeuralExpert customizes a neural network
specifically for your problem. The NeuralExpert
places buttons directly on the breadboard to
explain how this specific neural network works
and allow you to modify or test it.
Finally, choose a level of complexity for your
neural network. Simple networks will typically
train faster and produce better results. More
complex networks can be useful for learning more
complex relationships within the data.
20
NeuralBuilder
NeuralBuilder
Topology-based neural network creator.
The icon-based user interface of NeuroSolutions
provides unprecedented design flexibility for
constructing a neural network. This level of
access normally requires that you have a
substantial amount of neural network knowledge.
The NeuralBuilder eliminates this requirement by
hiding the complexities of the network and
simplifying the design process down to an easy,
step-by-step procedure.
Simply select a neural model, and then answer a
few questions about its configuration parameters.
The NeuralBuilder will compute default values for
each parameter based on your input data. These
default values will typically produce a network
that is capable of solving your problem. However,
the real power of NeuroSolutions is the level of
access provided to you for parameter
optimization. All of the parameters of the
constructed network can be completely customized.
21
NeuralBuilder
Step 1
Models supported by NeuralBuilder
Model Selection

• Multilayer Perceptron
• Generalized Feedforward Network
• Modular Neural Network
• Jordan/Elman Network
• Principal Component Analysis Network
• Radial Basis Function Network
• Self-Organizing Feature Map Network
• Time-Lag Recurrent Network
• Recurrent Network
• Generalized Regression Network
• Probabilistic Network
• CANFIS Network (Fuzzy Logic)
• Support Vector Machine

Select desired neural model from list provided.
Respective descriptions appear under list box.
22
NeuralBuilder
Step 3
Step 2
Cross Validation Data
Training Data
Select the testing file(s) used for cross
validation. This data can either be extracted
from the training set or read from separate files.
Select the training file(s), tagging the input
and desired data. Included are facilities for
data prediction and symbol translation.
23
NeuralBuilder
Step 5
Step 4
Layer Configuration
Topology Configuration
Specify the number of processing elements,
activation function, gradient search method, and
learning rate for each network layer.
Specify global parameters relating to the
networks topology for the selected neural model.
This can be as simple as specifying the number of
layers.
24
NeuralBuilder
Step 6
Step 7
Simulation Control
Probe Configuration
Specify probes to visualize the data.At each of
the five most common network points, you can
choose the probe which is most appropriate for
the data at that point.
Specify a stop criterion for the training. This
can be based on the number of iterations and/or
the error in either the training set or test set.
25
Data Manager
Data Manager
Data management and analysis tool.
Do you want to manage multiple datasets and
analyze them in a simple user interface? The Data
Manager allows you to easily analyze, preprocess
and partition data. Also included is the ability
to plot the data and view the results in the same
screen. The datasets are saved within one data
project, enabling simple file organization and
providing a user-friendly interface to manipulate
your datasets.
Simply open the data set and perform several
different type of analyses or create your neural
network right in the Data Manager. The Data
Manager is directly tied into the NeuralBuilder
which will compute default values for each
parameter based on your input data. These default
values will typically produce a network that is
capable of solving your problem. However, the
real power of NeuroSolutions is the level of
access provided to you for parameter
optimization. All of the parameters of the
constructed network can be completely
customized.
26
Data Manager
Step 2
Step 1
Analyze Data
Opening Data
Select Analyze Data to choose from several
different analyses functions.
Select Open Data File on the interface.
27
Data Manager
Step 4
Step 3
Partition Data
Preprocess Data
Select Partition Data on the interface to select
different options of segmenting the dataset.
Select Preprocess Data on the interface to choose
from many preprocessing features.
28
Data Manager
Step 6
Step 5
Manage Datasets
Plots
Select Manage Datasets on the interface to select
many options for data management .
Select Plot on the interface to plot the data in
a Time Series Plot or X-Y Scatter Plot.
29
Data Manager
Step 7
Build Neural Model
Select Build Neural Models on the interface to
begin creating your neural model in
NeuroSolutions.
The NeuralBuilder will compute default values for
each parameter based on your input data. These
default values will typically produce a network
that is capable of solving your problem. All of
the parameters of the constructed network can be
completely customized.
30
Advanced Features
Genetic Optimization
Sensitivity Analysis
After training a neural network, you may want to
know the effect that each of the network inputs
is having on the network output. Sensitivity
analysis is a method for extracting the cause and
effect relationship between the inputs and
outputs of the network. The input channels that
produce low sensitivity values can be considered
insignificant and can most often be removed from
the network. This will reduce the size of the
network, which in turn reduces the complexity and
the training time. Furthermore, this may also
improve the network performance.
All levels of NeuroSolutions Users level and
above include Genetic Optimization. Genetic
Optimization allows you to optimize virtually any
parameter in a neural network to produce the
lowest error. For example, the number of hidden
units, the learning rates, and the input
selection can all be optimized to improve the
network performance. Individual weights used in
the neural network can even be updated through
Genetic Optimization as an alternative to
traditional training methods.
31
Advanced Features
Code Generation
Dynamic Link Libraries (DLLs)
The Professional level generates ANSI-compatible
C source code for any network, including
learning. This allows a simulation prototyped
within the GUI to be run on other hardware
platforms. In addition, NeuroSolutions networks
can be integrated into your own applications.
The Developers level allows you to integrate your
own algorithms into NeuroSolutions through
dynamic link libraries (DLL). Every GUI component
implements a function belonging to
NeuroSolutions Simulation Protocol. Developers
can add components by simply writing
ANSI-compatible C functions that conform to this
protocol.
32
Advanced Features
Macros
OLE Automation
Embedded in NeuroSolutions is a comprehensive
macro language, which allows the user to record a
sequence of operations and store them as a
program. Any action that can be performed using
the mouse and keyboard can be duplicated with a
macro statement. This powerful feature gives the
user unprecedented flexibility in constructing,
editing, and running neural networks. When
running the NeuroSolutions demos, keep in mind
that they were constructed entirely with macros.
NeuroSolutions is a fully compliant OLE
Automation Server. This means that NeuroSolutions
can receive control messages from OLE Automation
Controllers, such as Visual Basic, Microsoft
Excel, Microsoft Access, and Delphi. Writing a
fully-functioning VB program is as simple as
recording a NeuroSolutions macro, clicking on the
convert to VB button, and pasting the converted
VB code into the desired VB application. A VB
application could be written to set a networks
parameters, run the network, then retrieve the
networks output.
33
There are six different levels of NeuroSolutions.
Select any of the levels below for a description,
or use the arrows to advance through them one at
a time.
Educator
Users
Consultants
Professional
Information is also available for the following
options and add-on products.
Developers Lite
Developers
Source Code License
NeuroSolutions For Excel
Custom Solution Wizard
34
Educator Level
Unrestricted Topologies Multilayer perceptions
(MLPs) Generalized feedforward networks Up
to 50 inputs/neurons per layer
Learning Paradigms Backpropagation

• Competitive Advantage
• Double-precision calculations
• 32-bit code
• Faster simulations
• Icon-based graphical user interface
• Extensive probing capabilities
• Easy neural network creation with the
• NeuralExpert and the NeuralBuilder

35
User Level
Unrestricted Topologies All topologies of the
Educator Modular networks Jordan-Elman
networks Self Organizing Feature Map
networks Radial Basis Function networks
Fuzzy Logic networks Support Vector Machine
networks Up to 500 inputs/neurons per layer
Additional Features Genetic optimization of
neural network parameters and weights.

• Learning Paradigms
• Backpropagation
• All search methods of Educator level
• Conjugate Gradient
• Levenberg-Marquardt
• Unsupervised Learning
• Hebbian
• Ojas
• Sangers
• Competitive
• Kohonen

Competitive Advantage More neurons per
layer More neural models to choose from
More unsupervised learning rules
36
Consultants Level
Unrestricted Topologies All topologies of the
Users Hopfield networks Time Delay Neural
networks Time-Lag Recurrent networks
User-defined network topologies Over 90
components to build from A virtually infinite
number of possible networks
Learning Paradigms All paradigms of Users
Recurrent backpropagation Backpropagation
through time
Competitive Advantage Unlimited
inputs/outputs/neurons per layer Modular
design allowing user-defined network
topologies Dynamic systems modeling
Time-Lag Recurrent networks
37
Professional Level
Unrestricted Topologies All topologies of the
Consultants
Learning Paradigms All paradigms of Consultants
Additional Features ANSI C Source Code
generation for Visual C Borland compilers
Embed networks into your own applications
Train networks on faster computers (Code
generation for Unix requires Source Code License.)
38
Developers Lite Level
Unrestricted Topologies All topologies of the
Consultants
Learning Paradigms All paradigms of Consultants
Additional Features User-defined dynamic link
libraries Customized neural components
Nonlinearities Interconnection matrices
Gradient search procedures Error
criteria Unsupervised learning rules
Memory structures Customized input
Customized output Customized parameter
scheduling
39
Developers Level
Unrestricted Topologies All topologies of the
Consultants
Learning Paradigms All paradigms of Consultants
Additional Features All additional features of
Developers Lite All additional features of
Professional
40
Source Code License
The Professional and Developers levels of
NeuroSolutions allow you to generate
ANSI-compatible C source code for the networks
you create with the graphical user interface. The
generated code links against an object library
which contains the implementations for the neural
components. Pre-compiled libraries are included
for Visual C (6.0 7.0) and Borland C
Builder (3.0 or higher). In order to compile the
generated code on another platform such as UNIX,
or on another Windows compiler, you would need to
purchase the Source Code License. Included with
the license is the source code for the entire
object library, enabling you to compile this
library for your particular platform/compiler and
link it with the generated code.
41
NeuroSolutions for Excel

• Unrestricted Topologies
• All topologies of the licensed level of
  NeuroSolutions
• Learning Paradigms
• All learning paradigms of the licensed version
  of NeuroSolutions
• Additional Features
• Data Preprocessing and Analysis
• Visual Data Selection
• Training and testing from within Microsoft
  Excel
• Leave-N-Out Training
• Parameter Optimization
• Sensitivity Analysis
• Automated Report Generation
• Custom Batch Creation / Execution

42
Custom Solution Wizard

• Unrestricted Topologies
• All topologies of the licensed level of
  NeuroSolutions and the Custom Solution Wizard
• Learning Paradigms
• All learning paradigms of the licensed version
  of NeuroSolutions and the Custom Solution
  Wizard
• Additional Features
• Generates and compiles a Dynamic Link Library
  (DLL) for any NeuroSolutions neural network
• Supports both recall and learning networks
  (Developers level)
• Allows you to easily embed a neural network
  into your own application developed with
• Visual Basic
• Microsoft Excel
• Microsoft Access
• Visual C
• Active Server Pages (ASP web pages)
• TradingSolutions
• NeuroSolutions for Matlab

43
NeuroSolutions for Excel
NeuroSolutions for Excel is a revolutionary
product which benefits both the beginner and
advanced neural network developer. For the
beginner, NeuroSolutions for Excel offers visual
data selection, one step training and testing,
and automated report generation. For the
advanced user, NeuroSolutions for Excel offers
the ability to perform parameter optimization,
run batch experiments, and create custom batch
experiments programmatically. The best part is
that all of these tasks can be performed without
ever leaving Microsoft Excel. NeuroSolutions for
Excel is organized into the seven modules listed
below.
Select any of the following modules for more
information or use the arrows to step through
them all.
Preprocess Data
Create Data Files
Analyze Data
Train Network
Tag Data
Test Network
Create/Open Network
44
Preprocess Data Module
The Preprocess Data module allows you to easily
apply various preprocessing techniques to your
raw data to prepare it for input into a neural
network. You can also create your own custom
Preprocess Data batches by calling built-in
NeuroSolutions functions and/or writing Visual
Basic code. These custom batches can then be run
from the NeuroSolutions for Excel menu from
within Microsoft Excel. The following Preprocess
Data operations are built into NeuroSolutions for
Excel

• Difference Computes the difference or percent
  difference along a column of data.
• Randomize Rows Randomly arranges the rows of
  data within the active worksheet and writes the
  result to
• a new worksheet.
• Sample Creates a new worksheet made up of every
  Nth row of data within the active worksheet.
• Moving Average Computes the moving average of a
  column using the chosen window length.
• Translate Symbolic Columns Translates columns
  that have been tagged as symbol.
• Insert Column Labels Inserts a row of column
  labels into the first row of the active
  worksheet.
• Clean Data Cleans the data by replacing blank
  cells, error codes, and/or user-defined text with
  an interpolated value, the column average, a
  random value, or the closest value in a column.
• Shift The input data is adjusted to either move
  the inputs back by a specified shift value to do
  predictions or move the inputs forward to lead
  your desired output.
• Encode Two Class Column The selected column of
  data is checked to verify that there are two
  classes contained within the column and is
  then encoded into another column. The data to be
  encoded can be textual or numeric, but must be
  translated to only numeric, integer codes. The
  encoded column will be written in the first
  empty column in the dataset.

45
Analyze Data Module
The Analyze Data module provides you with useful
information about your data. The operations
available in this module can be used during the
preprocessing stage of neural network design or
to analyze the network output. You can also
create your own custom Analyze Data batches by
calling built-in NeuroSolutions functions and/or
writing Visual Basic code. These custom batches
can then be run from the NeuroSolutions for Excel
menu from within Microsoft Excel. The following
Analyze Data operations are built into
NeuroSolutions for Excel

• Correlation Computes the correlation between
  each of the columns of data on the active
  worksheet.
• Time Series Plot Creates a Time Series Plot of
  the selected columns.
• XY Scatter Plot Creates an XY Scatter Plot of
  the selected columns.
• Histogram Computes the histogram of a selected
  column of data.
• Summary Statistics Computes various statistics
  for a selected column of data.
• Trend Accuracy Computes the trend accuracy of
  the selected columns.

46
Tag Data Module
The Tag Data module provides a simple graphical
method for tagging portions of your data as
Training Input, Training Desired, Cross
Validation Input, Cross Validation Desired,
Testing Input, Testing Desired, and Production
Input. This module also provides powerful
autotag methods. You can also create your own
custom Tag Data batches by calling built-in
NeuroSolutions functions and/or writing Visual
Basic code. These custom batches can then be run
from the NeuroSolutions for Excel menu from
within Microsoft Excel. The following Tag Data
operations are built into NeuroSolutions for
Excel

• Column(s) As Input Tags the selected column(s)
  of data as Input.
• Column(s) As Desired Tags the selected
  column(s) of data as Desired.
• Column(s) As Symbol Tags the selected column(s)
  of data as Symbol.
• Row(s) As Training Tags the selected row(s) of
  data as Training.
• Row(s) As Cross Validation Tags the selected
  row(s) of data as Cross Validation.
• Row(s) As Testing Tags the selected row(s) of
  data as Testing.
• Row(s) As Production Tags the selected row(s)
  of data as Production.
• All Columns As Input Tags all columns as Input.
• All Non-Numeric Columns As Symbol Tags all
  non-numeric columns as symbol.
• All Rows As Training Tags all rows as Training.
• Rows By Percentages Tags the rows of data
  within the active worksheet as Training, Cross
  Validation, and Testing according to
  user-defined percentages.
• Clear Tags Allows you to clear any existing
  tag.
• Clear Column Tag Clears the tag(s) of the
  selected column(s).
• Clear Symbol Tag Clears the symbol tag for the
  selected column(s).
• Clear Row Tag Clear the tag(s) of the selected
  row(s).
• Clear All Tags Clears all of the tags on the
  active worksheet.
• Select Cross-Section Allows you to
  automatically select any existing cross-section.
• Refresh Tag Formatting Refreshes the tag
  formatting.

47
Create/Open Network Module
The Create/Open Network module allows you to
create a NeuroSolutions breadboard from scratch
through the use of the NeuralBuilder utility or
open an existing NeuroSolutions breadboard. You
can also create your own custom Create Network
batches by calling built-in NeuroSolutions
functions and/or writing Visual Basic code.
These custom batches can then be run from the
NeuroSolutions for Excel menu from within
Microsoft Excel. The following Create/Open
Network operations are built into NeuroSolutions
for Excel

• New Classification Network Creates a new
  NeuroSolutions breadboard with typical elements
  used for a classification problem.
• New Function Approximation Network Creates a new
  NeuroSolutions breadboard with typical elements
  used for a classification problem.
• New Custom Network Starts the NeuralBuilder
  which guides you step-by-step through the
  creation of a new NeuroSolutions breadboard.
• Open Opens an existing NeuroSolutions
  breadboard.
• Close Closes the active NeuroSolutions
  breadboard.
• Save Saves the active NeuroSolutions
  breadboard.
• Save As Allows you to save the active
  NeuroSolutions breadboard to a user- specified
  location.
• Load Best Weights Loads the best weights for the
  active network.
• Tile Excel/NS Horizontally tiles NeuroSolutions
  and Microsoft Excel.

48
Create Data Files Module
The Create Data Files module creates tab
delimited ASCII files for each tagged
cross-section. You can also create your own
custom Create Data Files batches by calling
built-in NeuroSolutions functions and/or writing
Visual Basic code. These custom batches can then
be run from the NeuroSolutions for Excel menu
from within Microsoft Excel. The following
Create Data Files operations are built into
NeuroSolutions for Excel

• All Files Creates data files for all tagged
  cross-sections within the active worksheet.
• Training Files Creates Training Input and
  Training Desired files from the correspondingly
  tagged cross-sections within the active
  worksheet.
• Cross Validation Files Creates Cross Validation
  Input and Cross Validation Desired files from the
  correspondingly tagged cross-sections within
  the active worksheet.
• Testing Files Creates Testing Input and Testing
  Desired files from the correspondingly tagged
  data cross-sections within the active
  worksheet.
• Production Input File Creates Production Input
  file from the correspondingly tagged data
  cross-section within the active worksheet.
• View Data File Allows you to view (in Notepad) a
  data file that was created for the active
  worksheet.
• Delete Data Files Deletes all of the files
  previously created for the active worksheet.

49
Train Network Module
The Train Network module gives you the ability to
train a network once, multiple times with
different random initial conditions, and multiple
times while varying a network parameter. This
powerful module permits you to easily find the
optimum network for a particular problem. You
can also create your own custom Train Network
batches by calling built-in NeuroSolutions
functions and/or writing Visual Basic code.
These custom batches can then be run from the
NeuroSolutions for Excel menu from within
Microsoft Excel. The following Train Network
operations are built into NeuroSolutions for
Excel

• Train Trains the active NeuroSolutions
  breadboard one time and creates a report of
  the results.
• Train N Times Trains the active NeuroSolutions
  breadboard N times with different random
  initial conditions and creates report of the
  results.
• Vary a Parameter Trains the active
  NeuroSolutions breadboard N times for each value
  of a network parameter as the parameter is
  varied from a user defined starting value by a
  user-defined increment for a user defined number
  of variations.
• Leave-N-Out Training Trains the network multiple
  times, each time omitting a different subset of
  the data and using that subset for testing.
  The outputs from each tested subset are
  combined into one testing report and the model
  is trained one additional time using all of
  the data.
• Train Genetic Trains the active NeuroSolutions
  breadboard while genetically optimizing the
  choice of inputs and parameter values to
  achieve the best model.

50
Test Network Module
The Test Network module can be used to test a
network after training has been completed. In
testing the network, various performance measures
are computed. This module also allows you to
perform sensitivity analysis on the network. You
can also create your own custom Test Network
batches by calling built in NeuroSolutions
functions and/or writing Visual Basic code.
These custom batches can then be run from the
NeuroSolutions for Excel menu from within
Microsoft Excel. The following Test Network
operations are built into NeuroSolutions for
Excel

• Test Tests the active NeuroSolutions breadboard
  on the chosen data set and creates a report
  of the results.
• Sensitivity About the Mean Performs sensitivity
  analysis on the chosen data set.

51
NeuroSolutions
Reviews
NeuroSolutions is the most powerful and flexible
neural network simulator available for Windows,
but dont just take our word for it. The menus
below link to reviews of NeuroSolutions from
several top AI magazines.
Select any of the following reviews to view them
now.Each of the reviews will appear in a
separate document viewing window.
IEEE Spectrum Review
PC AI Magazine Review
PC AI Magazine Vendors Forum Review
The following review is also available for
viewing on the web. This review is only
available while connected to the internet.
EE Times Review
You can find out more about these magazines by
visiting their web sites.These links are only
available while connected to the internet.
PC AI Magazine
IEEE Spectrum
EE Times
52
NeuroSolutions
Sample Customer Applications
NeuroSolutions can be used to design neural
networks to solve many different types of
real-world problems.The topics in the menu below
link to application summaries written by
NeuroDimension customers on how they are using
NeuroSolutions software to solve and study real
life problems in a variety of fields. These
summaries are just a sample of the wide variety
of fields to which NeuroSolutions can be applied.
Select any of the following topics to view a
customer application summary.Each of the
summaries will appear in a separate document
viewing window.
Medicine
Psychology
Social Sciences
Finance
Marketing
Image Processing
Education
Management
Instrumentation
Flow Control
Signals
Theory Generation
53
Custom Solution Wizard
The Custom Solution Wizard is a tool that will
take an existing neural network created with
NeuroSolutions and automatically generate and
compile a Dynamic Link Library (DLL), allowing
you to easily incorporate your neural network
solutions into your own Visual Basic, Microsoft
Excel, Microsoft Access, or Visual C
applications. You can even use your DLL from the
internet in an Active Server Page or directly
from TradingSolutions! These generated neural
network DLLs can be used to respond to data based
on weights files you created within
NeuroSolutions. DLLs generated with the
Developers Level of the Custom Solution Wizard
also support learning, allowing your programs to
adapt to new data at runtime. The process of
communicating with the generated DLL is made
extremely simple through the use of the freely
distributable NeuroSolutions Object Library DLL.
This ActiveX DLL provides a simple protocol for
sending data and receiving the neural network
responses. By simply adding the NeuroSolutions
Object Library DLL to the references list of your
development environment, all of its methods and
properties will immediately be available to your
program. Embedding a custom neural network into
your application could not be any easier!
Select any of the following topics for more
information or use the arrows to step through
them all.
Wizard Walkthrough
Visual Basic Example
Feature Summary
54
The Custom Solution Wizard can be launched from
NeuroSolutions or directly from your Windows
start menu or desktop. When you launch it, just
indicate whether you would like to use the
breadboard that is currently active in
NeuroSolutions or open a breadboard you have
created previously.
If you select not to use the active breadboard,
you will be asked to select which breadboard to
use to generate your neural network DLL. The
Custom Solution Wizard can be used to generate
DLLs for breadboards like those created with the
Neural Wizard, as well as breadboards based on
your own designs.
55
The Custom Solution Wizard can generate a project
shell showing you how to use your neural network
DLL in Visual Basic, Visual C, Microsoft Excel,
or Microsoft Access. You can even use your DLL
from the internet in an Active Server Page,
TradingSolutions or in NeuroSolutions for Matlab!
Finally, select the location where your neural
network DLL and weights files should be placed.
Press Finish and the wizard will create a neural
network DLL and project shell for your custom
solution. Its that easy!
56
Sample Code
The following Visual Basic code demonstrates just
how easy it is to use the neural network DLL
generated by the Custom Solution Wizard. This
example creates a new NSRecallNetwork object,
sends it the XOr input data, and gets the network
response, all in just a few lines of code!
'Create the input data array. 'You can use
existing data files, databases,
spreadsheets, 'hardware devices or anything else
with data! Dim inputData(0 To 1, 0 To 3) As
Variant inputData(0, 0) 0! inputData(0, 1)
0! inputData(0, 2) 1! inputData(0, 3)
1! inputData(1, 0) 0! inputData(1, 1)
1! inputData(1, 2) 0! inputData(1, 3)
1! 'Create a new NeuroSolutions NSRecallNetwork
object. 'Visual Basic automatically knows how to
use this object'when you add the Object Library
to your reference list! Dim nn As New
NSRecallNetwork
'Set path to the generated recall network DLL.
nn.dllPathName "c\XOrBreadboard.dll" 'Set
path to weights file from previous training
sessions. nn.loadWeights "c\XOrBreadboard.nsw"
'Send input data to the network DLL.
nn.inputData inputData 'Get the network
response to this data. Dim outputData As Variant
outputData nn.getResponse 'Display the
output in a message box. 'You can use the network
response data in the same 'way you would use any
other data in your application! MsgBox "Output 1
" outputData(0,0) _ ", Output 2 "
outputData(1,0) _ ", Output 3 "
outputData(2,0) _ ", Output 4 "
outputData(3,0)
57
(No Transcript)
58
NeuroSolutions for MATLAB
The NeuroSolutions for MATLAB neural network
toolbox is a valuable addition to MATLAB's
technical computing capabilities allowing users
to leverage the power of NeuroSolutions inside
MATLAB. The toolbox features 15 neural models, 5
learning algorithms and a host of useful
utilities integrated in an easy-to-use interface,
which requires next to no knowledge of neural
networks to begin using the product. It allows
you to concentrate on solving your problem using
neural networks without having to spend many
taxing hours perusing neural network literature
and developing the algorithms yourself. The
toolbox is also integrated with NeuroSolutions.
This enables users to build custom networks in
NeuroSolutions, generate DLLs for those networks
using the Custom Solution Wizard and then use
those neural network DLLs inside MATLAB using the
NeuroSolutions for MATLAB interface. The three
products are available as a suite at a discounted
price.
Select any of the following topics for more
information or use the arrows to step through
them all.
Feature Summary
Product Tour
59
Product Tour CREATING A NEURAL NETWORK The
easiest way to create a neural network using
NeuroSolutions for MATLAB is to type the
following command within the MATLAB interface.
gtgt mynet nsnn The preceding command
creates the default network, a one hidden layer
Multi-Layer Perceptron (MLP), which is the most
popular neural network among engineers and
researchers worldwide. All the settings for the
network are set to well-researched defaults,
putting the neural network in a good-to-go
state after entering just one simple command.
USING SMART DEFAULTS Other parameters that
depend on your actual data are set when the data
is passed to the train function (nsTrain). For
example, the ideal number of neurons (processing
elements) in the hidden layer of the neural
network is computed from the data using a
proprietary formula. Thus, the intricacies
involved in setting up a neural network are
automatically taken care of, allowing the user to
concentrate on solving the problem at hand.
Here you have seen how to create the default
MLP network. Many other neural networks and
learning algorithms are available within
NeuroSolutions for MATLAB.
60
TRAINING YOUR NEURAL NETWORK The following
command trains the neural network with your
data. gtgt mynet nsTrain (mynet, x, y) where,
x is the input data and y is the desired data.
Cross validation can be performed without any
additional effort by passing the cross validation
data to the train function as well. gtgt mynet
nsTrain (mynet, x, y, cv_x, cv_y) where cv_x
is the cross-validation input data and cv_y is
the cross-validation desired data.
MONITORING TRAINING The learning curve and the
output and desired plots can be seen with ease
after training by setting their respective
parameters to true. gtgt mynet nsnn gtgt
mynet.learningCurve true gtgt
mynet.outputAndDesired true gtgt mynet
nsTrain (mynet, inputData, desiredData)
Output and Desired
Learning Curve
61
TESTING THE NEURAL NETWORK After training, the
performance of the neural network model can be
evaluated on a new out-of-sample testing data
set. gtgt z_out, perf nsTest (mynet, z_in,
z_desired) gtgt perf perf mse 0.7316 nmse
0.1728 correlation 0.9095 percent_error
13.1862 where z_in and z_desired represent the
testing input and desired data respectively.
z_out represents the output that the network
actually produced when tested with z_in. The
variable perf stores indicators comparing the
network output z_out with the desired output
z_desired.
UTILIZING THE NEURAL NETWORK Once you have
created the network, trained and tested it to
your satisfaction, the neural network is ready to
be utilized in practice with production data.
gtgt p_out nsProduction (mynet, p_in) where
p_in is the production input data and p_out is
the network output for the production input data.

62
EASY-TO-USE INTERFACE The NeuroSolutions neural
network (nsnn) object created has many different
parameters that can be edited. When the nsnn
object variable name is displayed in the MATLAB
command line, every parameter is displayed with
short comments that immediately explain the
function of that parameter. If there is any
doubt in your mind as to what a parameter is
actually used for, the help for any of the
parameters can be obtained simply by typing
help after the parameter name. gtgt
mynet.updateMethod.help
SAMPLE HELP DISPLAY
UPDATEMETHOD This parameter indicates the update
method used to effect weight updates during
training. It takes values batch, online and
custom. batch - Weights are updated after
every epoch of presentation. online - Weights
are updated after every pattern/exemplar. custom
- Weights are updated after every x number of
patterns/exemplars of pattern representation. The
number of patterns x after which a weight update
is effected is indicated in the
numExemplarsPerUpdate parameter. Example mynet
nsnn mynet.updateMethod batch mynet
nsnn mynet.updateMethod custom mynet.numExem
plarsPerUpdate 100 shortcut um
63
Easy-to-use Interface The functionality
available in the toolbox is integrated in an
easy-to-use interface that can be utilized by
users with next to no knowledge of neural
networks. Users who are familiar with MATLAB
would be able to pick up and use the entire
package within a few minutes. 15 Neural
Models The toolbox features several variants of
the following neural models Multi-Layer
Perceptron Generalized Feed Forward network
Modular neural network Support Vector Machine
Partially Recurrent neural network Fully
Recurrent neural network Time-Lag Recurrent
neural network
5 Learning Algorithms The following 5 learning
algorithms are featured, including the powerful
ConjugateGradient method Step Momentum
Quickprop Delta-Bar-Delta Conjugate Gradient
Useful Utilities The following utilities are
also included Symbolic data translation Image
flattening utility Performance indicators
Symbolic data translation allows for using
textual data as inputs to a neural network. The
image flattening utility flattens an image into a
single row of data, so that it can be fed into a
neural network. The performance indicators
function reveals how well the neural network has
trained with statistical indicators.
64
TradingSolutions is a full-featured financial
product that incorporates the power of neural
networks to help you to track and predict
financial market data. Easily create predictions
using wizards that handle the complexities of
neural model selection and data preprocessing.
View your data using built-in charts and
spreadsheets. Analyze your predictions and
trading signals for profitability using
comprehensive analysis tools. Make your financial
data work for you with TradingSolutions.
Screenshot
Features
Click here to return to the ND main menu.
65
Easy-to-use Data Management InterfaceView your
financial data in a familiar Explorer-like tree
structure. Import data from a wide variety of
sources including other financial packages,
ASCII text files, or even data collected from the
internet.
Powerful Financial CalculationsAnalyze and
preprocess your data using one of over 150
built-in functions or write your own functions
using the powerful formula entry system. Import
and export custom functions from TradingSolutions
to share with other users.
Built-in Chart and Spreadsheet ToolsVisualize
any financial data, indicator, trading system, or
prediction using built-in chart and spreadsheet
tools. Display data as lines, bars, HLOC, and
candlesticks along with buy/sell signals
generated from your trading systems.
Trading System Editing and Analysis ToolsUse
data from financial calculations and neural
predictions to create market-trading signals.
Automatically analyze your signals to determine
their profit potential based on historical data.
Advanced Neural Network TechnologyPredict
financial market trends using advanced neural
network technology. Use intuitive wizards to
create financials models which easily outperform
traditional modeling techniques.
Integration with other NeuroDimension
ProductsImport your TradingSolutions data
directly into NeuroSolutions for use in custom
breadboards. Then use Custom Solution Wizard to
integrate your neural topologies with
TradingSolutions.
Flexible Genetic OptimizationOptimize neural
network parameters using state-of-the-art genetic
algorithm technology. This enables you to build
more accurate neural models with the touch of a
button.
Animated Demonstrations and Step-by-step
TutorialsQuickly learn how to use
TradingSolutions by viewing animated how-to
demonstrations. Then, after viewing the
demonstrations, let the step-by-step tutorials
walk you through several real-world examples.
TradingSolutions uses the Genetic Libraryfor its
genetic optimization!
Click here for product information.
66
(No Transcript)
67
Genetic Server and Genetic Library provide a
general purpose API for genetic algorithm design.
Genetic Server is an ActiveX component that can
be used to easily build a custom genetic
application in Visual Basic. Genetic Library is a
C library that can be used for building custom
genetic applications in C.There are no
royalties for distributing applications built
with the ActiveX component or the library.
Click here to return to the ND main menu.
68
What are Genetic Algorithms? Genetic algorithms
are general-purpose search algorithms based upon
the principles of evolution observed in nature.
Genetic algorithms combine selection, crossover,
and mutation operators with the goal of finding
the best solution to a problem. Genetic
algorithms search for this optimal solution until
a specified termination criterion is met.The
solution to a problem is called a chromosome. A
chromosome is made up of a collection of genes
which are simply the parameters to be optimized.
A genetic algorithm creates an initial population
(a collection of chromosomes), evaluates this
population, then evolves the population through
multiple generations (using the genetic operators
discussed above) in the search for a good
solution for the problem at hand. Genetic
algorithms can be applied to a wide variety of
optimization problems such as scheduling,
computer games, stock market trading, medical,
adaptive control, transportation, the traveling
salesmen problem