Presentation is loading. Please wait.

Presentation is loading. Please wait.

Neural Optimization of Evolutionary Algorithm Strategy Parameters Hiral Patel.

Similar presentations


Presentation on theme: "Neural Optimization of Evolutionary Algorithm Strategy Parameters Hiral Patel."— Presentation transcript:

1 Neural Optimization of Evolutionary Algorithm Strategy Parameters Hiral Patel

2 Outline Why optimize parameters of an EA? Why optimize parameters of an EA? Why use neural networks? Why use neural networks? What has been done so far in this field? What has been done so far in this field? Experimental Model Experimental Model Preliminary Results and Conclusion Preliminary Results and Conclusion Questions Questions

3 Why optimize parameters of an EA? Faster convergence Faster convergence Better overall results Better overall results Avoid premature convergence Avoid premature convergence

4 Why use neural networks? Ability to learn Ability to learn Adaptability Adaptability Pattern recognition Pattern recognition Faster then using another EA Faster then using another EA

5 What has been done so far in this field? Machine Learning primarily used to optimize ES and EP Machine Learning primarily used to optimize ES and EP Optimized mutation operators Optimized mutation operators Little has been done to optimize GA parameters Little has been done to optimize GA parameters

6 Experimental Model Outline Neural Network Basics Neural Network Basics Hebbian Learning Hebbian Learning Parameters of the Genetic Algorithm to be optimized Parameters of the Genetic Algorithm to be optimized Neural Network Inputs Neural Network Inputs

7 Neural Network Basics Weight update algorithm w q1 (k) w q2 (k) w qn (k) v q (k) d q (k) g()=f’() y q (k) b q (k) bias Vector input signal x(k)  R n  1 Deviation of activation function Synaptic weights Desired neuron response Neuron response (output) Sigmoid activation function f() Adapted from: Ham, M. H., Kostanic, I Principles of Neurocomputing for Science and Engineering, McGraw-Hilll, NY, 2001

8 Hebbian Learning Unsupervised learning Unsupervised learning Time-dependent Time-dependent Learning signal and Forgetting factor Learning signal and Forgetting factor

9 Hebb Learning for single neuron Standard Hebbian learning rule { ,  } x1x1 xnxn w0w0 w1w1 wnwn v f(v)y Adapted from: Ham, M. H., Kostanic, I Principles of Neurocomputing for Science and Engineering, McGraw-Hilll, NY, 2001 x0x0

10 Parameters of the Genetic Algorithm to be optimized Crossover Probability Crossover Probability Crossover Cell Divider Crossover Cell Divider Cell Crossover Probability Cell Crossover Probability Mutation Probability Mutation Probability Mutation Cell Divider Mutation Cell Divider Cell Mutation Probability Cell Mutation Probability Bit Mutation Probability Bit Mutation Probability

11 Neural Network Inputs Current Parameter Values Current Parameter Values Variance Variance Mean Mean Max fitness Max fitness Average bit changes for crossover Average bit changes for crossover Constant parameters of the GA Constant parameters of the GA

12 Preliminary Results Tests run with Knapsack problem with dataset 3, pop. size 800, rep. size 1600 Tests run with Knapsack problem with dataset 3, pop. size 800, rep. size 1600 Learning Signal and Forgetting factor are not yet optimal enough to suggest better performance with NN Learning Signal and Forgetting factor are not yet optimal enough to suggest better performance with NN

13 Output for 1600 generations

14 Probabilities for 1600 generations

15 Conclusion It may be possible to get better performance out of a Neural Optimized EA as long as the (unsupervised) Neural Network is able to adapt to the changes quickly and to recognize local minima. It may be possible to get better performance out of a Neural Optimized EA as long as the (unsupervised) Neural Network is able to adapt to the changes quickly and to recognize local minima.

16 Possible Future Work ES to optimize parameters, use a SOM to do feature extraction of the optimized parameter values, use the SOM output as codebook vectors for LVQ network and then classify the output of the original ES, use the classifications to perform supervised training of Levenberg- Marquardt Backpropagation network to form rule set. ES to optimize parameters, use a SOM to do feature extraction of the optimized parameter values, use the SOM output as codebook vectors for LVQ network and then classify the output of the original ES, use the classifications to perform supervised training of Levenberg- Marquardt Backpropagation network to form rule set.

17 Question ?


Download ppt "Neural Optimization of Evolutionary Algorithm Strategy Parameters Hiral Patel."

Similar presentations


Ads by Google