Tensor flow for Android

TensorFlow is the most popular framework for Machine Learning, and we can use it in Android. We explain how TensorFlow works and show an example.

Transcript

1. Sorting Hat with TensorFlow Jorge Juan Barroso Carmona jorge@karumi.com @flipper83 +JorgeJBarroso Android Expert Android GDE
2. Testing for android & iOS. Trainings Architecture, Patterns and principles for Android & iOS. Mastering Git. Advanced mobile development. Testing for android & iOS. Architecture, Patterns and principles for Android & iOS. Companies For Everybody
3. The best mobile app performance monitoring tool.
4. Adam Tornhill “Of course it is happening inside your mobile, Harry,but why on earth should that mean that it is not real?” J.K Rowling and Jorge Barroso, Harry Potter and the deathly Hallows
5. Sorting Hat
6. The Sorting Hat is one of the cleverest enchanted objects most witches and wizards will ever meet. It literally contains the intelligence of the four founders.
7. www.tiny.cc/sortingHat
8. https://github.com/flipper83/SortingHatAndroid
9. www.tiny.cc/sortingHat
10. Neural Nets
11. Deep Learning
12. Classification Discovery
13. x1 x2 x3 x4 w1 w2 w3 w4 w5 w6 w7 w8 w9
14. x1 x2 w1 w2 b1 y = x1*w1 + x2*w2 + b1 Activation function TRY sigmoid functions instead lineal
15. b1
16. x1 x2 x3 x4 Training Hidden Layers 1/2(error)2 Accuracy Error
17. x1 x2 x3 x4 Validation (backpropagation) Hidden Layers Accuracy Error
18. Convolutional Networks
23. https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/
26. Adam Tornhill Multidimensional array Computational Graph
27. Tensors Transform information Into dimensional matrix
28. Tensors
29. SHAPE RANK TYPE
30. 2D 480x320 float32
31. Variables Are the way the have To give importance to data structures
32. Variables
33. MINIMIZATION RESTORE SAVING
34. Neural Nets Are the structures for building Core deep learning models
35. Neural Nets
36. ACTIVATION FUNCTION HIDDEN LAYERS OPTIMIZERS
37. ReLU [50,100,50] Gradient descent
38. The Sorting Hat
39. INCEPTION 2.Retrain the model
40. 1.Gather images
41. /retrained_graph_optimized.pb /retrained_labels.txt 4.Import the model in Android application
42. 5. Write some code
43. classifier = TensorFlowImageClassifier.create( getAssets(), MODEL_FILE, LABEL_FILE, INPUT_SIZE, IMAGE_MEAN, IMAGE_STD, INPUT_NAME, OUTPUT_NAME); private static final float IMAGE_STD = 128; protected static final int INPUT_SIZE = 299; protected static final int IMAGE_MEAN = 128; private static final String MODEL_FILE = "file:/// android_asset/retrained_graph_optimized.pb"; private static final String LABEL_FILE = "file:/// android_asset/retrained_labels.txt";
44. final List<Classifier.Recognition> results = classifier.recognizeImage(croppedBitmap); for (Classifier.Recognition result : results) { Log.d(result.getTitle() + " " + result.getConfidence()); }
46. Bibliography They’re the cracks! Peter Warden. Tensorflow mobile for Poets 2016 Hello word with Tensorflow Introduction to Tensorflow. Gema Parreño medium 2016 Pottermore. JKRowling https://rubenlopezg.wordpress.com/2014/05/07/que-es-y-como- funciona-deep-learning/ https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/ https://cloud.google.com/blog/big-data/2016/07/understanding- neural-networks-with-tensorflow-playground http://nilhcem.com/android/custom-tensorflow-classifier http://www.asimovinstitute.org/neural-network-zoo
47. Find me I am very social! jorge@karumi.com @flipper83 Questions?