Slide 1
Slide 1 text
Mobile, AI and Tensorflow
Supriya Srivatsa
Slide 2
Slide 2 text
No content
Slide 3
Slide 3 text
No content
Slide 4
Slide 4 text
NEURAL NETWORKS
Human anatomy inspired learning network.
Slide 5
Slide 5 text
Neural Networks
Slide 6
Slide 6 text
Neural Networks
Slide 7
Slide 7 text
Neural Network – A Peek Inside
Slide 8
Slide 8 text
Deep Neural Network
Slide 9
Slide 9 text
PREDICTION AND INFERENCE
How it works today. How it shall work tomorrow.
Slide 10
Slide 10 text
“Transfer to Infer” Approach
Slide 11
Slide 11 text
Why On-Device Prediction
• Data Privacy
• Poor internet connections
• Questionable user experience
Slide 12
Slide 12 text
To The Rescue…
Slide 13
Slide 13 text
TensorFlow
• Tensor: N Dimensional Arrays
• Open source software library for numerical
computation using data flow graphs.
Slide 14
Slide 14 text
TensorFlow – Data Flow Graphs
• Nodes represent mathematical functions
• Edges represent tensors.
Slide 15
Slide 15 text
Tensorflow – “Deferred Execution” Model
• Graph first. Computation Afterward.
import tensorflow as tf
x = tf.constant(10)
y = tf.Variable(x + 5)
print(y)
Slide 16
Slide 16 text
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))
Slide 17
Slide 17 text
No content
Slide 18
Slide 18 text
No content
Slide 19
Slide 19 text
Packaging the App and the Model
Slide 20
Slide 20 text
QUANTIZATION
Compress. And Compress More.
Slide 21
Slide 21 text
Quantization
• Round it up
• Transform: round_weights
• Compression rates: ~8% => ~70%
• Shrink down node names
• Transform: obfuscate_names
• Eight bit calculations
Slide 22
Slide 22 text
Quantization - Eight Bit Calculations
Slide 23
Slide 23 text
Quantization - Eight Bit Calculations
Slide 24
Slide 24 text
No content
Slide 25
Slide 25 text
IMPLEMENTATION
Code Away! ☺
Slide 26
Slide 26 text
Implementation
build.gradle
buildscript {
repositories {
jcenter()
}
dependencies {
classpath 'com.android.tools.build:gradle:2.3.0'
}
}
Slide 27
Slide 27 text
Implementation
1. Load
2. Feed
3. Run
4. Fetch
Slide 28
Slide 28 text
Implementation
1. Load the model
2. Feed in the input
3. Run the model
4. Fetch the output
TensorFlowInferenceInterface inferenceInterface =
new TensorFlowInferenceInterface(assetManager, modelFile);
Slide 29
Slide 29 text
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);
Slide 30
Slide 30 text
Implementation
1. Load the model
2. Feed in the input
3. Run the model
4. Fetch the output
inferenceInterface.run(outputNames);
Slide 31
Slide 31 text
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);
Slide 32
Slide 32 text
APPLICATIONS
Awesomeness.
Slide 33
Slide 33 text
Google Translate
Slide 34
Slide 34 text
No content
Slide 35
Slide 35 text
No content