SER594 Lecture 07

Human Computer Interaction
Neural Net
(201903)

Transcript

1. SER594 Human-Computer Interaction Lecture 07 Neural Net Javier Gonzalez-Sanchez, PhD javiergs@asu.edu javiergs.engineering.asu.edu Office Hours: By appointment
2. Integration 2 Agent Integration Agent Agent System Multimodal
3. Continuous Dimensional Model
4. Continuous Dimensional Model -++ --- --+ +-- ++- -+- --+
5. Continuous Dimensional Model
6. What about? • Skin Conductance • Pupil Size • Heart Rate
7. Pattern Recognition
8. Tools
9. Neural Network | Neuron input output
10. Neural Network | Neuron X1 X2 X3 x4
11. Neural Network | Neuron X1 X2 X3 x4 • Add • Normalize
12. Neural Network | Neuron X1=1 X2=0.5 X3=0.5 X4=-1 • Add • Normalize 0.73
13. Neural Network | Weights X1=1 X2=0.5 X3=0.5 X4=-1 0.37 W1=0.5 W2=1 W3=1 W4=2 0.18 -3.7 W1=0.5 W2=1 W3=-10
14. Neural Network | Weights
15. Neural Network And, How to train the model? Weights are important
16. Neural Network X1 X2 X3 X4 W1 W2 W3 W4 sig (W1*X1 + W2*X2 + W3*X3 + W4*X4)
17. Neural Network X1 X2 X3 X4 W1,1 W2,1 W3,1 W4,1 sig (W11*X1 + W21*X2 + W31*X3 + W41*X4) W1,2 W2,2 W3,2 W4,2 sig (W12*X1 + W22*X2 + W32*X3 + W42*X4) W [i][j] = W[weight][neuron]
18. Neural Network X1 X2 X3 X4 sig (W111*X1 + W121*X2 + W131*X3 + W141*X4) W1,1,2 W1,2,2 W1,3,2 W1,4,2 W [k][i][j] = W[layer][weight] [neuron] sig ( ) sig (W112*X1 + W122*X2 + W132*X3 + W142*X4) sig ( ) W2,1,1 W2,2,1 W2,1,2 W2,2,2 layer_1 layer_2
19. Neural Network | Weights input output
20. Cost Function
21. Back propagation • Partial derivative of the cost function with respect to any weight in the network. • We want to see how the function changes as we let just one of those variables change while holding all the others constant. • Black-box
22. Example
23. Hand writing • 60,000 of training data • 10,000 of test data • Black white digit images of 28X28 pixels • Each pixel contains a number from 0-255 showing the gray scale, 0 while and 255 black. http://ramok.tech/tag/java-neural-networks-example/
24. Hand writing • Input (X) – array 784 numbers 0-255 • 60,000 of these • Output - array of 10 elements (probabilities) • More layer better it is, but the training it will also be much slower. http://ramok.tech/tag/java-neural-networks-example/
25. Hand writing public void train(Integer trainData, Integer testFieldValue) { initSparkSession(); List<LabeledImage> labeledImages = IdxReader.loadData(trainData); List<LabeledImage> testLabeledImages = IdxReader.loadTestData(testFieldValue); Dataset<Row> train = sparkSession.createDataFrame(labeledImages, LabeledImage.class).checkpoint(); Dataset<Row> test = sparkSession.createDataFrame(testLabeledImages,LabeledImage.class).checkpoint(); //in=28x28=784, hidden layers (128,64), out=10 int[] layers = new int[]{784, 128, 64, 10}; MultilayerPerceptronClassifier trainer = new MultilayerPerceptronClassifier() .setLayers(layers) .setBlockSize(128) .setSeed(1234L) .setMaxIter(100); model = trainer.fit(train); // evalOnTest(test); }
26. Homework http://ramok.tech/tag/java-neural-networks-example/
27. Homework Recognize states
28. SER594 – Human Computer Interaction Javier Gonzalez-Sanchez javiergs@asu.edu Spring 2019 Disclaimer. These slides can only be used as study material for the SER594 course at ASU. They cannot be distributed or used for another purpose.