C. Benatti, 3/15/2012, Slide 1 GA/ICA Workshop Carla Benatti 3/15/2012
C. Benatti, 3/15/2012, Slide 2
Proposed Thesis Project• Tuning a Beam Line
– Model/design of system provides nominal values for tune
– Operators adjust each element individually to optimize tune
– Slow process, room for improvement
• Tuning Algorithm and Optimizer– Develop new, fast, tuning algorithm– Using neural networks, genetic
algorithms possibly– Model Independent Analysis
• Benchmark code at ReA3– Design experiment to test optimizer– Compare results with tuning “by hand”– User friendly application, possibly GUI
L051 L054 L057 L061LB006LB004
LB source, L-line at ReA3
COSY Envelope tracking calculation
C. Benatti, 3/15/2012, Slide 3
Artificial Neural Network (ANN)• Neural Network Summary
– Attempts to simulate the functionality of the brain in a mathematical model
– Ideal for modeling complex relationships between inputs and outputs as a “black box” solver
– Ability to learn, discern patterns, model nonlinear data
– Reliability of prediction– Many different models already
developed for finding local and global minimum for optimization
• Neural Network Programming– Neuron receives weighted input– If above threshold, generates output
through nonlinear function– Connecting single neurons together
creates a neural network– Learning, training: get ANN to give a
desired output, supervised or unsupervised learning (GA example)
x1
x2
xN
w1
w2
wN
y
∑1=
_ )(=N
iii bxwφy
y = Outputw = Weightsx = Inputsb = Thresholdφ = Non-linear Function
Neuron
Input layerHidden layer(s)
Output layerx1
x2
xN
11
22
m
k
NeuronwN
3
Multilayer Perceptron
• Basic ANN example
• Hierarchical structure
• Feed-forward network
Perceptron
C. Benatti, 3/15/2012, Slide 4
Genetic Algorithms• Machine learning technique• Effective tool to deal with complex
problems by evolving creative and competitive solutions
• Genetic Algorithms search for the optimal set of weights, thresholds for neurons
• Crossover is the most used search operator in Genetic Programming
Iterate
Terminate
End
Reproduction
http://www.ai-junkie.com/ann/evolved/nnt7.html
(0.3, -0.8, -0.2, 0.6, 0.1, -0.1, 0.4, 0.5)
Elitism
(0.3, -0.8, -0.2, 0.6, 0.1, -0.1, 0.4, 0.5)(0.7, 0.4, -0.9, 0.3, -0.2, 0.5, -0.4, 0.1)
(0.7, 0.4, -0.9, 0.6, 0.1, -0.1, 0.4, 0.5)
Parents
Crossover
Mutation(0.7, 0.4, -0.9, 0.6, 0.1, -0.3, 0.4, 0.5)
Genetic Modification Examples
C. Benatti, 3/15/2012, Slide 5
SmartSweepers Tutorial Code
• NeuralNet.m• NeuralNet_CalculateOutput.m• Genetic_Algorithm.m
http://www.ai-junkie.com/ann/
smart sweepers.exe
Best Fitness
Average Fitness
smart sweepers.exe