extreme learning machine for multi-categories...

55
tu-logo ur-logo Outline Extreme Learning Machine for Multi- Categories Classification Applications Hai-Jun Rong 1,2 , Guang-Bin Huang 1 and Yew-Soon Ong 2 1 School of Electrical and Electronic Engineering 2 School of Computer Engineering Nanyang Technological University Nanyang Avenue, Singapore 639798 E-mail: {hjrong, egbhuang, asysong}@ntu.edu.sg IEEE World Congress on Computational Intelligence Hong Kong, June 1-6 2008 ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Upload: others

Post on 10-Nov-2020

7 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Outline

Extreme Learning Machine for Multi-Categories Classification Applications

Hai-Jun Rong1,2, Guang-Bin Huang1 and Yew-Soon Ong2

1School of Electrical and Electronic Engineering

2School of Computer EngineeringNanyang Technological University

Nanyang Avenue, Singapore 639798E-mail: {hjrong, egbhuang, asysong}@ntu.edu.sg

IEEE World Congress on Computational IntelligenceHong Kong, June 1-6 2008

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 2: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Outline

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 3: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Outline

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 4: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Outline

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 5: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Outline

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 6: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Outline

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 7: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 8: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Feedforward Neural Networks with Additive Nodes

Figure 1: Feedforward Network Architecture: additive hiddennodes

Output of hidden nodes

G(ai , bi , x) = g(ai · x + bi ) (1)

ai : the weight vector connecting the i th hiddennode and the input nodes.bi : the threshold of the i th hidden node.

Output of SLFNs

fL(x) =LX

i=1

βi G(ai , bi , x) (2)

βi : the weight vector connecting the i th hiddennode and the output nodes.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 9: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Feedforward Neural Networks with Additive Nodes

Figure 1: Feedforward Network Architecture: additive hiddennodes

Output of hidden nodes

G(ai , bi , x) = g(ai · x + bi ) (1)

ai : the weight vector connecting the i th hiddennode and the input nodes.bi : the threshold of the i th hidden node.

Output of SLFNs

fL(x) =LX

i=1

βi G(ai , bi , x) (2)

βi : the weight vector connecting the i th hiddennode and the output nodes.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 10: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Feedforward Neural Networks with Additive Nodes

Figure 1: Feedforward Network Architecture: additive hiddennodes

Output of hidden nodes

G(ai , bi , x) = g(ai · x + bi ) (1)

ai : the weight vector connecting the i th hiddennode and the input nodes.bi : the threshold of the i th hidden node.

Output of SLFNs

fL(x) =LX

i=1

βi G(ai , bi , x) (2)

βi : the weight vector connecting the i th hiddennode and the output nodes.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 11: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Feedforward Neural Networks with RBF Nodes

Figure 2: Feedforward Network Architecture: RBF hiddennodes

Output of hidden nodes

G(ai , bi , x) = g (bi‖x − ai‖) (3)

ai : the center of the i th hidden node.bi : the impact factor of the i th hidden node.

Output of SLFNs

fL(x) =LX

i=1

βi G(ai , bi , x) (4)

βi : the weight vector connecting the i th hiddennode and the output nodes.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 12: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Feedforward Neural Networks with RBF Nodes

Figure 2: Feedforward Network Architecture: RBF hiddennodes

Output of hidden nodes

G(ai , bi , x) = g (bi‖x − ai‖) (3)

ai : the center of the i th hidden node.bi : the impact factor of the i th hidden node.

Output of SLFNs

fL(x) =LX

i=1

βi G(ai , bi , x) (4)

βi : the weight vector connecting the i th hiddennode and the output nodes.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 13: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Feedforward Neural Networks with RBF Nodes

Figure 2: Feedforward Network Architecture: RBF hiddennodes

Output of hidden nodes

G(ai , bi , x) = g (bi‖x − ai‖) (3)

ai : the center of the i th hidden node.bi : the impact factor of the i th hidden node.

Output of SLFNs

fL(x) =LX

i=1

βi G(ai , bi , x) (4)

βi : the weight vector connecting the i th hiddennode and the output nodes.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 14: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 15: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Function Approximation of Neural Networks

Figure 3: Feedforward Network Architecture.

Learning Model

For N arbitrary distinct samples(xi , ti ) ∈ Rn × Rm , SLFNs with Lhidden nodes and activation functiong(x) are mathematically modeled as

fL(xj ) = oj , j = 1, · · · , N (5)

Cost function: E =PN

j=1

‚‚‚oj − tj‚‚‚

2.

The target is to minimize the costfunction E by adjusting the networkparameters: βi , ai , bi .

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 16: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Function Approximation of Neural Networks

Figure 3: Feedforward Network Architecture.

Learning Model

For N arbitrary distinct samples(xi , ti ) ∈ Rn × Rm , SLFNs with Lhidden nodes and activation functiong(x) are mathematically modeled as

fL(xj ) = oj , j = 1, · · · , N (5)

Cost function: E =PN

j=1

‚‚‚oj − tj‚‚‚

2.

The target is to minimize the costfunction E by adjusting the networkparameters: βi , ai , bi .

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 17: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Function Approximation of Neural Networks

Figure 3: Feedforward Network Architecture.

Learning Model

For N arbitrary distinct samples(xi , ti ) ∈ Rn × Rm , SLFNs with Lhidden nodes and activation functiong(x) are mathematically modeled as

fL(xj ) = oj , j = 1, · · · , N (5)

Cost function: E =PN

j=1

‚‚‚oj − tj‚‚‚

2.

The target is to minimize the costfunction E by adjusting the networkparameters: βi , ai , bi .

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 18: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Function Approximation of Neural Networks

Figure 3: Feedforward Network Architecture.

Learning Model

For N arbitrary distinct samples(xi , ti ) ∈ Rn × Rm , SLFNs with Lhidden nodes and activation functiong(x) are mathematically modeled as

fL(xj ) = oj , j = 1, · · · , N (5)

Cost function: E =PN

j=1

‚‚‚oj − tj‚‚‚

2.

The target is to minimize the costfunction E by adjusting the networkparameters: βi , ai , bi .

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 19: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Learning Algorithms of Neural Networks

Figure 4: Feedforward Network Architecture.

Learning Methods

Many learning methodsmainly based ongradient-descent/iterativeapproaches have beendeveloped over the pasttwo decades.

Back-Propagation (BP)and its variants are mostpopular.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 20: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Learning Algorithms of Neural Networks

Figure 4: Feedforward Network Architecture.

Learning Methods

Many learning methodsmainly based ongradient-descent/iterativeapproaches have beendeveloped over the pasttwo decades.

Back-Propagation (BP)and its variants are mostpopular.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 21: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Learning Algorithms of Neural Networks

Figure 4: Feedforward Network Architecture.

Learning Methods

Many learning methodsmainly based ongradient-descent/iterativeapproaches have beendeveloped over the pasttwo decades.

Back-Propagation (BP)and its variants are mostpopular.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 22: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Advantagnes and Disadvantages

PopularityWidely used in various applications: regression,classification, etc.

LimitationsUsually different learning algorithms used in differentSLFNs architectures.Some parameters have to be tuned mannually.Overfitting.Local minima.Time-consuming.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 23: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

SLFN ModelsLearning Methods

Advantagnes and Disadvantages

PopularityWidely used in various applications: regression,classification, etc.

LimitationsUsually different learning algorithms used in differentSLFNs architectures.Some parameters have to be tuned mannually.Overfitting.Local minima.Time-consuming.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 24: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 25: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Figure 5: Feedforward Network Architecture: any type ofG(ai , bi , x).

New Learning Theory

If continuous target function f (x) can beapproximated by SLFNs with adjustable hiddennodes then the hidden node parameters of suchSLFNs needn’t be tuned. Instead, all thesehidden node parameters can be randomlygenerated without the knowledge of the trainingdata. Given any nonconstant piecewisecontinuous function g, for any continuous targetfunction f and any randomly generatedsequence {(ai , bi )

Li=1},

limL→∞

‖f (x) − fL(x)‖ = 0

holds with probability one if βi is chosen tominimize the ‖f (x) − fL(x)‖, i = 1, · · · , L.

G.-B. Huang, et al., “Universal Approximation Using Incremental Constructive Feedforward Networks with Random

Hidden Nodes,” IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Convex Incremental Learning Machine,” Neurocomputing, vol. 70, pp. 3056-3062, 2007.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 26: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Figure 5: Feedforward Network Architecture: any type ofG(ai , bi , x).

New Learning Theory

If continuous target function f (x) can beapproximated by SLFNs with adjustable hiddennodes then the hidden node parameters of suchSLFNs needn’t be tuned. Instead, all thesehidden node parameters can be randomlygenerated without the knowledge of the trainingdata. Given any nonconstant piecewisecontinuous function g, for any continuous targetfunction f and any randomly generatedsequence {(ai , bi )

Li=1},

limL→∞

‖f (x) − fL(x)‖ = 0

holds with probability one if βi is chosen tominimize the ‖f (x) − fL(x)‖, i = 1, · · · , L.

G.-B. Huang, et al., “Universal Approximation Using Incremental Constructive Feedforward Networks with Random

Hidden Nodes,” IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Convex Incremental Learning Machine,” Neurocomputing, vol. 70, pp. 3056-3062, 2007.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 27: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Unified Learning Platform

Figure 6: Feedforward Network Architecture: any type ofG(ai , bi , x).

Mathematical Model

For N arbitrary distinct samples(xi , ti ) ∈ Rn × Rm , standard SLFNswith L hidden nodes and output functiong(x) are mathematically modeled as

LXi=1

βi G(ai , bi , xj ) = tj , j = 1, · · · , N

(6)

(ai , bi ): Hidden node parameters.

βi : the weight vector connecting the i thhidden node and the output node.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 28: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Unified Learning Platform

Figure 6: Feedforward Network Architecture: any type ofG(ai , bi , x).

Mathematical Model

For N arbitrary distinct samples(xi , ti ) ∈ Rn × Rm , standard SLFNswith L hidden nodes and output functiong(x) are mathematically modeled as

LXi=1

βi G(ai , bi , xj ) = tj , j = 1, · · · , N

(6)

(ai , bi ): Hidden node parameters.

βi : the weight vector connecting the i thhidden node and the output node.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 29: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Mathematical ModelPLi=1 βi G(ai , bi , xj ) = tj , j = 1, · · · , N is equivalent to Hβ = T, where

H(a1, · · · , aL, b1, · · · , bL, x1, · · · , xN )

=

2664G(a1, b1, x1) · · · G(aL, bL, x1)

.

.

. · · ·...

G(a1, b1, xN ) · · · G(aL, bL, xN )

3775N×L

(7)

β =

26664βT

1...

βLT

37775L×m

and T =

26664tT1...

tTN

37775N×m

(8)

H is called the hidden layer output matrix of the neural network; the i th column of H is the output of the i thhidden node with respect to inputs x1, x2, · · · , xN .

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 30: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Outline

1 Neural NetworksSingle-Hidden Layer Feedforward Networks (SLFNs)Conventional Learning Algorithms of SLFNs

2 Extreme Learning MachineUnified Learning PlatformELM Algorithm

3 ELM for Multi-Categories Classification Problems

4 Performance Evaluations

5 Summary

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 31: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Three-Step Learning Model

Given a training set ℵ = {(xi , ti)|xi ∈ Rn, ti ∈ Rm, i = 1, · · · , N}, activationfunction g, and the number of hidden nodes L,

1 Assign randomly hidden node parameters (ai , bi), i = 1, · · · , L.2 Calculate the hidden layer output matrix H.3 Calculate the output weight β: β = H†T.

where H† is the Moore-Penrose generalized inverse of hidden layer outputmatrix H.

Source Codes of ELM

http://www.ntu.edu.sg/home/egbhuang/

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 32: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Three-Step Learning Model

Given a training set ℵ = {(xi , ti)|xi ∈ Rn, ti ∈ Rm, i = 1, · · · , N}, activationfunction g, and the number of hidden nodes L,

1 Assign randomly hidden node parameters (ai , bi), i = 1, · · · , L.2 Calculate the hidden layer output matrix H.3 Calculate the output weight β: β = H†T.

where H† is the Moore-Penrose generalized inverse of hidden layer outputmatrix H.

Source Codes of ELM

http://www.ntu.edu.sg/home/egbhuang/

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 33: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Three-Step Learning Model

Given a training set ℵ = {(xi , ti)|xi ∈ Rn, ti ∈ Rm, i = 1, · · · , N}, activationfunction g, and the number of hidden nodes L,

1 Assign randomly hidden node parameters (ai , bi), i = 1, · · · , L.2 Calculate the hidden layer output matrix H.3 Calculate the output weight β: β = H†T.

where H† is the Moore-Penrose generalized inverse of hidden layer outputmatrix H.

Source Codes of ELM

http://www.ntu.edu.sg/home/egbhuang/

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 34: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Three-Step Learning Model

Given a training set ℵ = {(xi , ti)|xi ∈ Rn, ti ∈ Rm, i = 1, · · · , N}, activationfunction g, and the number of hidden nodes L,

1 Assign randomly hidden node parameters (ai , bi), i = 1, · · · , L.2 Calculate the hidden layer output matrix H.3 Calculate the output weight β: β = H†T.

where H† is the Moore-Penrose generalized inverse of hidden layer outputmatrix H.

Source Codes of ELM

http://www.ntu.edu.sg/home/egbhuang/

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 35: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

Extreme Learning Machine (ELM)

Three-Step Learning Model

Given a training set ℵ = {(xi , ti)|xi ∈ Rn, ti ∈ Rm, i = 1, · · · , N}, activationfunction g, and the number of hidden nodes L,

1 Assign randomly hidden node parameters (ai , bi), i = 1, · · · , L.2 Calculate the hidden layer output matrix H.3 Calculate the output weight β: β = H†T.

where H† is the Moore-Penrose generalized inverse of hidden layer outputmatrix H.

Source Codes of ELM

http://www.ntu.edu.sg/home/egbhuang/

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 36: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Unified Learning PlatformELM Algorithm

ELM Learning Algorithm

Salient Features

“Simple Math is Enough.” ELM is a simple tuning-free three-step algorithm.

The learning speed of ELM is extremely fast.

Unlike the conventional learning methods which MUST see the training data before generating the hiddennode parameters, ELM could generate the hidden node parameters before seeing the training data.

Unlike the traditional classic gradient-based learning algorithms which only work for differentiable activationfunctions, ELM works for all bounded nonconstant piecewise continuous activation functions includingnon-differential activation functions.

Unlike the traditional classic gradient-based learning algorithms facing several issues like local minima,improper learning rate and overfitting, etc, ELM tends to reach the solutions straightforward without suchtrivial issues.

The ELM learning algorithm looks much simpler than many learning algorithms: neural networks andsupport vector machines.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 37: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

ELM for Multi-Categories Classification Problems

Three basic methods

1 Single ELM classifier: m output nodes of ELM for m-class applications.We say x is in class l if output node l has the highest output value.

2 One-Against-All ELM (ELM-OAA): m-class classification problems areimplemented by m binary ELM classifiers, each of which is trainedindependently to classify one of the m pattern classes.

3 One-Against-One ELM (ELM-OAO): the m pattern classes are pairwisedecomposed into m(m − 1)/2 two different classes, and each of them istrained by one binary ELM classifier.

Exponential loss based decoding approach used in ELM-OAA and ELM-OAOE. L. Allwein, et al., “Reducing multiclass to binary: a unifying approach for margin classifiers,” Journal of Machine

Learning Research, vol. 1, pp. 113-141, 2001.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 38: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

ELM for Multi-Categories Classification Problems

Three basic methods

1 Single ELM classifier: m output nodes of ELM for m-class applications.We say x is in class l if output node l has the highest output value.

2 One-Against-All ELM (ELM-OAA): m-class classification problems areimplemented by m binary ELM classifiers, each of which is trainedindependently to classify one of the m pattern classes.

3 One-Against-One ELM (ELM-OAO): the m pattern classes are pairwisedecomposed into m(m − 1)/2 two different classes, and each of them istrained by one binary ELM classifier.

Exponential loss based decoding approach used in ELM-OAA and ELM-OAOE. L. Allwein, et al., “Reducing multiclass to binary: a unifying approach for margin classifiers,” Journal of Machine

Learning Research, vol. 1, pp. 113-141, 2001.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 39: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

ELM for Multi-Categories Classification Problems

Three basic methods

1 Single ELM classifier: m output nodes of ELM for m-class applications.We say x is in class l if output node l has the highest output value.

2 One-Against-All ELM (ELM-OAA): m-class classification problems areimplemented by m binary ELM classifiers, each of which is trainedindependently to classify one of the m pattern classes.

3 One-Against-One ELM (ELM-OAO): the m pattern classes are pairwisedecomposed into m(m − 1)/2 two different classes, and each of them istrained by one binary ELM classifier.

Exponential loss based decoding approach used in ELM-OAA and ELM-OAOE. L. Allwein, et al., “Reducing multiclass to binary: a unifying approach for margin classifiers,” Journal of Machine

Learning Research, vol. 1, pp. 113-141, 2001.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 40: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

ELM for Multi-Categories Classification Problems

Three basic methods

1 Single ELM classifier: m output nodes of ELM for m-class applications.We say x is in class l if output node l has the highest output value.

2 One-Against-All ELM (ELM-OAA): m-class classification problems areimplemented by m binary ELM classifiers, each of which is trainedindependently to classify one of the m pattern classes.

3 One-Against-One ELM (ELM-OAO): the m pattern classes are pairwisedecomposed into m(m − 1)/2 two different classes, and each of them istrained by one binary ELM classifier.

Exponential loss based decoding approach used in ELM-OAA and ELM-OAOE. L. Allwein, et al., “Reducing multiclass to binary: a unifying approach for margin classifiers,” Journal of Machine

Learning Research, vol. 1, pp. 113-141, 2001.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 41: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

ELM for Multi-Categories Classification Problems

Three basic methods

1 Single ELM classifier: m output nodes of ELM for m-class applications.We say x is in class l if output node l has the highest output value.

2 One-Against-All ELM (ELM-OAA): m-class classification problems areimplemented by m binary ELM classifiers, each of which is trainedindependently to classify one of the m pattern classes.

3 One-Against-One ELM (ELM-OAO): the m pattern classes are pairwisedecomposed into m(m − 1)/2 two different classes, and each of them istrained by one binary ELM classifier.

Exponential loss based decoding approach used in ELM-OAA and ELM-OAOE. L. Allwein, et al., “Reducing multiclass to binary: a unifying approach for margin classifiers,” Journal of Machine

Learning Research, vol. 1, pp. 113-141, 2001.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 42: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Table 1: Specification of Real-World Classification BenchmarkProblems

Type Datasets # Attributes # Classes # ObservationsTraining Testing

Type I Glass 9 6 110 104Vehicle 18 4 420 426Page Blocks 10 5 2,700 2,773Image Segmentation 19 7 1,100 1,210Satellite Image 36 6 4,400 2,035Shuttle 9 7 43,500 14,500DNA 180 3 1,000 1,186Optical Recognition of Handwritten Digits 64 10 3,823 1,797Pen-Based Recognition of Handwritten Digits 16 10 7,494 3,498

Type II Cancer 98 14 144 46Arrhythmia 279 16 300 152Letter Recognition 16 26 10,000 10,000

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 43: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Table 2: Comparison of network complexity of ELM-OAA, ELM andELM-OAO

Datasets ELM-OAA ELM ELM-OAOGlass 30 30 20Vehicle 90 110 50Page Blocks 130 160 50Image Segmentation 200 210 30Satellite Image 450 470 150Shuttle 260 340 80DNA 350 450 290Handwritten Optical 420 470 110

RecognitionHandwritten Pen-Based 650 690 190

RecognitionCancer 40 50 10Arrhythmia 50 70 20Letter Recognition 1100 1500 510

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 44: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Table 3: Comparison of testing accuracy of ELM-OAA, ELM andELM-OAO

Datasets ELM-OAA ELM ELM-OAOGlass 66.346 64.423 65.385Vehicle 79.513 79.489 81.378Page Blocks 95.873 95.761 95.397Image Segmentation 95.216 94.667 94.493Satellite Image 89.670 89.663 89.955Shuttle 99.667 99.715 99.581DNA 94.866 94.828 94.496Handwritten Optical 96.928 96.948 96.343

RecognitionHandwritten Pen-Based 98.294 98.302 97.819

RecognitionCancer 78.652 77.522 73.000Arrhythmia 65.441 65.395 61.770Letter Recognition 93.417 93.029 93.055

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 45: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

Table 4: Comparison of training and testing time (seconds) ofELM-OAA, ELM and ELM-OAO

Datasets ELM-OAA (s) ELM (s) ELM-OAO (s)Training Testing Training Testing Training Testing

Glass 0.0250 0.0617 0.0117 0.0773 0.0133 0.0219Vehicle 0.1508 0.0617 0.0680 0.0531 0.0547 0.0313Page Blocks 1.7125 0.5023 0.6641 0.2070 0.3422 0.2836Image Segmentation 3.3547 0.2953 0.6000 0.0906 0.1453 0.1102Satellite Image 43.838 1.6500 8.4320 0.2758 5.1039 1.0383Shuttle 151.88 7.4594 36.614 1.6516 30.395 4.8422DNA 7.0797 0.6227 5.0328 0.2359 4.7070 0.4578Handwritten Optical Recognition 39.002 2.0734 6.7414 0.3461 4.2766 2.5727Handwritten Pen-Based Recognition 140.17 5.3305 18.127 0.7742 17.923 7.3648Cancer 0.0491 0.0191 0.0072 0.0194 0.0469 0.0422Arrhythmia 0.1250 0.0586 0.0266 0.0273 0.1414 0.2281Letter Recognition 2295.6 73.006 76.066 1.9700 1429.5 279.57

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 46: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

Summary

ELM, ELM-OAO and ELM-OAA obtain similar testingaccuracies.ELM-OAO usually requires smaller number of hiddennodes than the single ELM classifier and ELM-OAA.The training time required by ELM-OAO is similar or lessthan ELM and ELM-OAA when the number of patternclasses is small (say, not larger than 10).However when the number of pattern classes is large (say,larger than 10), the training time cost by ELM-OAO is mostlikely higher than the single ELM classifier but still smallerthan ELM-OAA.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 47: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 48: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 49: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 50: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 51: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 52: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 53: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 54: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines

Page 55: Extreme Learning Machine for Multi-Categories ...pdfs.semanticscholar.org/42ea/aa55b18175123b5cb3acff2f2dc2e3e… · tu-logo ur-logo Outline Extreme Learning Machine for Multi-Categories

tu-logo

ur-logo

Neural NetworksELM

ELM for Multi-Categories Classification ProblemsPerformance Evaluations

Summary

References

References

G.-B. Huang, et al., “Universal Approximation Using Incremental Networks with Random HiddenComputational Nodes”, IEEE Transactions on Neural Networks, vol. 17, no. 4, pp. 879-892, 2006.

G.-B. Huang, et al., “Extreme Learning Machine: Theory and Applications,” Neurocomputing, vol. 70, pp.489-501, 2006.

G.-B. Huang, et al., “Convex Incremental Extreme Learning Machine,” Neurocomputing, vol. 70, pp.3056-3062, 2007.

M.-B. Li, et al., “Fully complex extreme learning machine,” Neurocomputing, vol. 68, pp. 306-314, 2005.

N.-Y. Liang, et al., “A Fast and Accurate On-line Sequential Learning Algorithm for Feedforward Networks,”IEEE Transactions on Neural Networks, vol. 17, no. 6, pp. 1411-1423, 2006.

Q.-Y. Zhu, et al., “Evolutionary Extreme Learning Machine”, Pattern Recognition, vol. 38, no. 10, pp.1759-1763, 2005.

G.-B. Huang, et al., “Can Threshold Networks Be Trained Directly?” IEEE Transactions on Circuits andSystems II, vol. 53, no. 3, pp. 187-191, 2006.

G.-B. Huang, et al., “Real-Time Learning Capability of Neural Networks”, IEEE Transactions on NeuralNetworks, vol. 17, no. 4, pp. 863-878, 2006.

ELM Web Portal: www.ntu.edu.sg/home/egbhuang Extreme Learning Machines