Transcript CSC 562 Final Presentation - Dave Pizzolo

Artificial Neural
Networks
CSC 562: Final Project
Dave Pizzolo
What is an Artificial Neural Network?
Definition
•An Artificial Neural Network (ANN) is a computer program that can
recognize patterns in a given collection of data and produce a model for that
data.
It resembles the brain in two respects:
1.Knowledge is acquired by the network through a learning process (trail
and error)
2.Interneuron connection strengths known as synaptic weights are used to
store the knowledge
Typical ANN Applications
1. Function Approximation
2. Classification
3. Time Series Prediction
4. Data Mining
Not this ANN
1 - Function Approximation
• You know your inputs and outputs, but do not know your function
• y = f(x) where
• x is a set of numeric inputs
• y is a set of numeric outputs
• f() is an unknown functional relationship between the input and the output
• The ANN must approximate f() in order to find the appropriate output for
each set of inputs
• Demo: Body Fat Percentage
2 - Classification
• Similar to the function approximation except that the output is a “class”,
thus they are discrete
• For example:
• Outputs = on or off
• Outputs = sick or healthy
• Demo: Optical Character Recognition (OCR)
• 0 = {1,0,0,0,0,0,0,0,0,0}
• 1 = {0,1,0,0,0,0,0,0,0,0}
• …
• 9 = {0,0,0,0,0,0,0,0,0,1}
3 - Time Series Prediction
• Time Series Prediction is similar to function approximation except that
time plays an important role
• In function approximation, information that is needed to create output is
contained in the input
• Image processing
• In time series prediction, information from the past is need to determine the
output
• Stock price prediction
• Demo: Predict Mackey Glass Chaotic Signal
• Chaos is a signal that has characteristics similar to randomness, but can be
predicted accurate in the short term (e.g. weather)
• Accurate predictions can be made only a few samples in advance
This Mackey
Not this Mackey
4 - Data Mining
• All three previous problems required a known output for each input
• In data mining, you do not know the answer ahead of time. You want to
extract data from the input
• Clustering
• Compression
• Principal Component Analysis
• This type of a network is called “unsupervised” because there is no
“teaching” signal
• Demo: Clustering with Competition
• Clustering 2D data into N different regions
• Use competitive (unsupervised) learning
NeuroDimension, Inc.
NeuroDimension, Inc.
•A software development company headquartered in Gainesville, Florida and founded
in 1991. It specializes in neural networks, adaptive systems, and genetic optimization
and makes software tools for developing and implementing these artificial
intelligence technologies. (http://en.wikipedia.org/wiki/NeuroDimension)
•Company website: http://www.nd.com/
Product
•NeuroSolutions: http://www.neurosolutions.com/
•30 minute video demo: http://www.neurosolutions.com/resources/videotour.html#
•FREE evaluation copy of software: http://www.nd.com/neurosolutions/download.html
•Sample data: http://www.nddownloads1.com/videos/NNAndNSIntroductionFiles.zip
Demo
Function Approximation
• NS Excel
• File --> Open --> BodyFat.xls
• NeuroSolutions --> Train Network --> Train
• Apply Production Dataset
Time Series Prediction
• File --> Open --> 2 TDNN CHAOS.NSB
• Highlight range
• Step Epoch
Data Mining
• File --> Open --> 48 CLUSTERING.NSB
• Reset
Classification
• Step Epoch
• File --> Open --> OCR.NSB
• Tools --> Customize --> control
Sample Problem
• Start
• Tools --> Neural Expert
• Reset + Zero Count
• Function Approximation --> Next
• Step Exem
• Browse --> MPGEvaluation.asc --> Next
• Select All (but MPG) --> Next
• Country --> Next
• Use Input File for Desired File --> Shuffle Data Files --> Next
• MPG --> Next
• Low --> Finish
• Start
• Testing --> Next --> Next --> Next --> Finish