Mobile, AI and TensorFlow

Transcript

1. Mobile, AI and Tensorflow Supriya Srivatsa
4. NEURAL NETWORKS Human anatomy inspired learning network.
5. Neural Networks
6. Neural Networks
7. Neural Network – A Peek Inside
8. Deep Neural Network
9. PREDICTION AND INFERENCE How it works today. How it shall work tomorrow.
10. “Transfer to Infer” Approach
11. Why On-Device Prediction • Data Privacy • Poor internet connections • Questionable user experience
12. To The Rescue…
13. TensorFlow • Tensor: N Dimensional Arrays • Open source software library for numerical computation using data flow graphs.
14. TensorFlow – Data Flow Graphs • Nodes represent mathematical functions • Edges represent tensors.
15. Tensorflow – “Deferred Execution” Model • Graph first. Computation Afterward. import tensorflow as tf x = tf.constant(10) y = tf.Variable(x + 5) print(y)
16. Tensorflow – “Deferred Execution” Model • Graph first. Computation Afterward. import tensorflow as tf x = tf.constant(10) y = tf.Variable(x + 5) model = tf.global_variables_initializer() with tf.Session() as session: session.run(model) print(session.run(y))
19. Packaging the App and the Model
20. QUANTIZATION Compress. And Compress More.
21. Quantization • Round it up • Transform: round_weights • Compression rates: ~8% => ~70% • Shrink down node names • Transform: obfuscate_names • Eight bit calculations
22. Quantization - Eight Bit Calculations
23. Quantization - Eight Bit Calculations
25. IMPLEMENTATION Code Away! ☺
26. Implementation build.gradle buildscript { repositories { jcenter() } dependencies { classpath 'com.android.tools.build:gradle:2.3.0' } }
27. Implementation 1. Load 2. Feed 3. Run 4. Fetch
28. Implementation 1. Load the model 2. Feed in the input 3. Run the model 4. Fetch the output TensorFlowInferenceInterface inferenceInterface = new TensorFlowInferenceInterface(assetManager, modelFile);
29. Implementation 1. Load the model 2. Feed in the input 3. Run the model 4. Fetch the output // feed(String s, float[] floats, long… longs) inferenceInterface.feed(inputName, floatValues, 1, inputSize, inputSize, 3);
30. Implementation 1. Load the model 2. Feed in the input 3. Run the model 4. Fetch the output inferenceInterface.run(outputNames);
31. Implementation 1. Load the model 2. Feed in the input 3. Run the model 4. Fetch the output // fetch(String s, float[] floats) inferenceInterface.fetch(outputName, outputs);
32. APPLICATIONS Awesomeness.
33. Google Translate