Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
SELF ORGANISING NETWORKS/MAPS (SOM) AND NEURAL NETWORK APPLICATIONS Outcomes Look at the theory of self-organisation. Other self-organising networks Look at examples of neural network applications Four requirements for SOM Weights in neuron must represent a class of pattern one neuron, one class Four requirements for SOM Inputs pattern presented to all neurons and each produces an output. Output: measure of the match between input pattern and pattern stored by neuron. Four requirements A competitive learning strategy selects neuron with largest response. Four requirements A method of reinforcing the largest response. Architecture The Kohonen network (named after Teuvo Kohonen from Finland) is a self-organising network Neurons are usually arranged on a 2dimensional grid Inputs are sent to all neurons There are no connections between neurons Architecture X Kohonen network Theory For a neuron output (j) is a weighted some: Where x is the input, w is the weights, net is the output of the neuron Four requirement-Kohonen networks True Euclidean distance and weighted sum Winner takes all Learning rule of Kohonen learning Output value The output of each neuron is the weighted sum There is no threshold or bias Input values and weights are normalized “Winner takes all” Initially the weights in each neuron are random Input values are sent to all the neurons The outputs of each neuron are compared The “winner” is the neuron with the largest output value Training Having found the winner, the weights of the winning neuron are adjusted Weights of neurons in a surrounding neighbourhood are also adjusted Neighbourhood X neighbourhood Kohonen network Training As training progresses the neighbourhood gets smaller Weights are adjusted according to the following formula: Weight adjustment The learning coefficient (alpha) starts with a value of 1 and gradually reduces to 0 This has the effect of making big changes to the weights initially, but no changes at the end The weights are adjusted so that they more closely resemble the input patterns Example A Kohonen network receives the input pattern 0.6 0.6 0.6. Two neurons in the network have weights 0.5 0.3 0.8 and -0.6 –0.5 0.6. Which neuron will have its weights adjusted and what will the new values of the weights be if the learning coefficient is 0.4? Answer The weighted sums are 0.96 and –0.3 so the first neuron wins. The weights become: w1 = 0.5 + 0.4 *(0.6 – 0.5) w1 = 0.5 + 0.4 * 0.1 = 0.5 + 0.04 = 0.54 w2 = 0.3 + 0.4 *(0.6 – 0.3) w2 = 0.3 + 0.4 * 0.3 = 0.3 + 0.12 = 0.42 w2 = 0.8 + 0.4 *(0.6 – 0.8) w2 = 0.8 - 0.4 * 0.2 = 0.8 - 0.08 = 0.72 Summary The Kohonen network is self-organising It uses unsupervised training All the neurons are connected to the input A winner takes all mechanism determines which neuron gets its weights adjusted Neurons in a neighbourhood also get adjusted Demonstration A demonstration of a Kohonen network learning has been taken from the following websites: http://www.patol.com/java/TSP/index.html http://www.samhill.co.uk/kohonen/index.htm Applications of Neural Networks ARTIFICIAL INTELLIGENCE TECHNIQUES Example Applications Analysis of data Classifying in EEG Pattern recognition in ECG EMG disease detection. Gueli N et al (2005) The influence of lifestyle on cardiovascular risk factors analysis using a neural network Archives of Gerontology and Geriatrics 40 157–172 To produce a model of risk facts in heart disease. MLP used The accuracy was relatively good for chlorestremia and triglyceremdia: Training phase around 99% Testing phase around 93% Not so good for HDL Subasi A (in press) Automatic recognition of alertness level from EEG by using neural network and wavelet coefficients Expert Systems with Applications xx (2004) 1–11 Electroencephalography (EEG) Recordings of electrical activity from the brain. Classifying operation Awake Drowsy Sleep MLP 15-23-3 Hidden layer – log-tanh function Output layer – log-sigmoid function Input is normalise to be within the range 0 to 1. Accuracy 95%+/-3% alert 93%+/-4% drowsy 92+/-5% sleep Feature were extracted and form the input to the network, from wavelets. Karsten Sternickel (2002) Automatic pattern recognition in ECG time series Computer Methods and Programs in Biomedicine 68 109–115 ECG – electrocardiographs – electrical signals from the heart. Wavelets again. Classification of patterns Patterns were spotted Abel et al (1996) Neural network analysis of the EMG interference pattern Med. Eng. Phys. Vol. 18, No. 1. pp. 12-l 7 EMG – Electromyography – muscle activity. Interference patterns are signals produce from various parts of a muscle- hard to see features. Applied neural network to EMG interference patterns. Classifying Nerve disease Muscle disease Controls Applied various different ways of presenting the pattern to the ANN. Good for less serve cases, serve cases can often be see by the clinician. Example Applications Wave prediction Controlling a vehicle Condition monitoring Wave prediction Raoa S, Mandal S(2005) Hindcasting of storm waves using neural networks Ocean Engineering 32 (2005) 667–684 MLP used to predict storm waves. 2:2:2 network Good correlation between ANN model and another model van de Ven P, Flanagan C, Toal D (in press) Neural network control of underwater vehicles Engineering Applications of Artificial Intelligence Semiautomous vehicle Control using ANN ANN replaces a mathematical model of the system. Silva et al (2000) THE ADAPTABILITY OF A TOOL WEAR MONITORING SYSTEM UNDER CHANGING CUTTING CONDITIONS Mechanical Systems and Signal Processing (2000) 14(2), 287-298 Modelling tool wear Combines ANN with other AI (Expert systems) Self organising Maps (SOM) and ART2 investigated SOM better for extracting the required information. Examples to try yourself A.1 Number recognition (ONR) http://www.generation5.org/jdk/demos.asp# neuralNetworks Details: http://www.generation5.org/content/2004/si mple_ocr.asp B.1 Kohonen Self Organising Example 1 http://www.generation5.org/jdk/demos.asp# neuralNetworks B.2 Kohonen 3D travelling salesman problem http://fbim.fhregensburg.de/~saj39122/jfroehl/diplom/eindex.html