1 OPTIMIZED INTERCONNECTIONS IN OPTIMIZED INTERCONNECTIONS IN PROBABILISTIC SELF-ORGANIZING LEARNING PROBABILISTIC SELF-ORGANIZING LEARNING Janusz Starzyk, Mingwei Ding, Haibo He Janusz Starzyk, Mingwei Ding, Haibo He School of EECS School of EECS Ohio University, Athens, OH Ohio University, Athens, OH February 14-16, 2005 February 14-16, 2005 Innsbruck, Austria Innsbruck, Austria
23
Embed
OPTIMIZED INTERCONNECTIONS IN PROBABILISTIC SELF-ORGANIZING LEARNING
OPTIMIZED INTERCONNECTIONS IN PROBABILISTIC SELF-ORGANIZING LEARNING. Janusz Starzyk, Mingwei Ding, Haibo He School of EECS Ohio University, Athens, OH February 14-16, 2005 Innsbruck, Austria. OUTLINE. Introduction Self-organizing neural network structure Optimal and fixed input weights - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
11
OPTIMIZED INTERCONNECTIONS IN OPTIMIZED INTERCONNECTIONS IN PROBABILISTIC SELF-ORGANIZING LEARNINGPROBABILISTIC SELF-ORGANIZING LEARNING
Janusz Starzyk, Mingwei Ding, Haibo He Janusz Starzyk, Mingwei Ding, Haibo He
School of EECSSchool of EECSOhio University, Athens, OHOhio University, Athens, OH
February 14-16, 2005February 14-16, 2005Innsbruck, AustriaInnsbruck, Austria
22
OUTLINEOUTLINE
• IntroductionIntroduction• Self-organizing neural network structure Self-organizing neural network structure • Optimal and fixed input weightsOptimal and fixed input weights
Local and sparse Local and sparse iinterconnectionsnterconnections
Online inputs selectionOnline inputs selectionFeature neurons and Feature neurons and
merging neuronsmerging neurons
44
SOLAR Hardware StructureSOLAR Hardware Structure
55
Neuron StructureNeuron Structure
66
Self-organizationSelf-organization
Each neuron has the ability to self-Each neuron has the ability to self-organize according to received organize according to received informationinformation• Functionality – chose internal arithmetic Functionality – chose internal arithmetic
Consider 3 inputs to a neuron with correct classification probabilities equal to pi
Estimated output probability pout for various input probabilities is as follows
1313
Binary weightingBinary weighting Simplified selection algorithm is desired for hardware Simplified selection algorithm is desired for hardware
implementationimplementation Choose 0 or 1 as the weights for all the connected inputsChoose 0 or 1 as the weights for all the connected inputs
This equation can be used to study the effect of adding This equation can be used to study the effect of adding or removing connections of different signal strengthor removing connections of different signal strength
n
P
n
Pn
n
S ij
ij
ij Cii
Ci
Cii
2
2
2
2
2
ˆ
1
ˆ1
ˆ
ˆ
n
P
n
S i
ˆ
ˆ
ˆ
1414
Binary weighting (cont’d)Binary weighting (cont’d) A stronger connection PA stronger connection Pmaxmax
a weaker connection Pa weaker connection Pmixmix Criteria for adding weaker Criteria for adding weaker
connectionconnection Pmix
Pmax 0.5
0.5
Pcomb
Gain of information for different Pmax and Pmix Threshold for adding a new connection
From previous results, selection criteria for From previous results, selection criteria for binary weighting can be established.binary weighting can be established.
Pmax=0.69 0.5
0.5Pcomb
Pmix>0.60
Pmix<0.60
0.5
Threshold for adding a weaker connection
Pmax=0.69
1616
Simulation resultsSimulation results
Case I: Prediction of financial performanceCase I: Prediction of financial performance• Based on S&P Research Insight DatabaseBased on S&P Research Insight Database• More than 10,000 companies includedMore than 10,000 companies included• Training and testing on 3-year periodTraining and testing on 3-year period• 192 features extracted192 features extracted• Kernel PCA used to reduce 192 features to 13~15Kernel PCA used to reduce 192 features to 13~15
Simulation results (cont’d)Simulation results (cont’d) Case II: Power quality disturbance classification problemCase II: Power quality disturbance classification problem
People spill out onto Madison Avenue in New People spill out onto Madison Avenue in New York after blackout hit. York after blackout hit.
Cars stopped about three-quarters of the way up Cars stopped about three-quarters of the way up the first hill of the Magnum XL200 ride at Cedar the first hill of the Magnum XL200 ride at Cedar
Point Amusement Park in Sandusky, Ohio. Point Amusement Park in Sandusky, Ohio. (15, August, 2003, CNN Report)(15, August, 2003, CNN Report)
THE COST:THE COST:According to the North According to the North
American Electric Reliability American Electric Reliability Council (NERC)Council (NERC)
1919
Formulation of the problem:Formulation of the problem:• Wavelet Multiresolution Analysis (MRA) is used Wavelet Multiresolution Analysis (MRA) is used
outage; sag with harmonic; swell with harmonicoutage; sag with harmonic; swell with harmonic
• Two hundred cases of each class were Two hundred cases of each class were generated for training and another 200 cases generated for training and another 200 cases were generated for testing. were generated for testing.
Reference [16]: T. K. A. Galil et. al, “Power Quality Disturbance Classification Using the Inductive Inference Approach, ” IEEE Transactions On Power Delivery, Vol.19, No.4, October, 2004
2121
XILINX
XILINX
VIRTEX XCV 1000
VIRTEX XCV 1000
Hardware DevelopmentHardware Development
2222
Conclusion Conclusion
Input selection strategyInput selection strategy Optimum weighting scheme theoryOptimum weighting scheme theory Simple binary weighting for Simple binary weighting for
practical usepractical use Searching criteria for useful Searching criteria for useful
connectionsconnections Application studyApplication study Hardware platform design Hardware platform design