TensorFlow - Learning Machines

Presented at #devfest #gdgcv GDG Cloud Vancouver and GDG Chicago with supporting code repo from: https://github.com/VikramTiwari/tensorflow-retrain-sample

Transcript

2. tf.name_scope Vikram Tiwari @Vikram_Tiwari !Expert
3. repository in “machine learning” category on GitHub #1
5. A multidimensional array. A graph of operations.
6. Multidimensional Arrays
7. Graph You define the graph in high-level language (Python) Graph is compiled and optimized Graph is executed (in part, or fully) on devices Nodes represent computations Data (tensors) flows between them
8. With special operations for deep learning Similar to NumPy Numpy TensorFlow add mul matmul … sum … add multiply matmul … reduce_sum … sigmoid relu …
9. Dataflow computation Python Program TensorFlow Graph https://www.tensorflow.org/get_started/get_started
10. Linear Regression y = mx + b input parameters result
11. Inference import tensorflow as tf x = tf.placeholder(shape=[None], dtype=tf.float32) m = tf.Variable(tf.random_normal([1])) b = tf.Variable(tf.random_normal([1])) y = m * x + b with tf.Session() as sess: sess.run(tf.initialize_all_variables()) print(sess.run(y, feed_dict={x: x_in})) Build the graph Prepare execution env Initialize variables Run the computation
12. Loss, Gradient Descent parameter #1 loss parameter #2
13. Putting it together loss = tf.reduce_mean(tf.square(y - y_train)) optimizer = tf.train.GradientDescentOptimizer(0.5) train = optimizer.minimize(loss) with tf.Session() as sess: sess.run(tf.global_variable_initalizer()) for i in range(1000): sess.run(train, feed_dict={x: x_data[i], y_label: y_data[i]}) Define a loss Create an optimizer Op to minimize the loss Iteratively run the training op Initialize variables
14. Playground http://playground.tensorflow.org/ Demo
15. TensorFlow Distributed Execution Engine CPU GPU Android iOS ... C++ Frontend Python Frontend ... Layers Estimator Models in a box Train and evaluate models Build models Keras Model Canned Estimators
16. Estimator inputs, labels fit() model_fn predict() evaluate() Sessions, Graphs, Loops export_savedmodel()
17. TensorBoard
18. Distributed By Kaewkasi at English Wikipedia, CC BY-SA 3.0, https://goo.gl/PD1K2N
19. TensorFlow Serving
20. Who uses it? #google Search Gmail Translate Maps Android Photos Speech YouTube Play … many others ... Production use in many areas: Internal TensorFlow launch Research use for: 100s of projects and papers
21. Transfer Learning https://github.com/VikramTiwari/tensorflow-retrain-sample Demo
22. Next Steps Tutorials and code tensorflow.org Intro to Deep Learning with TensorFlow Udacity class goo.gl/iHssII Stanford’s CS231n cs231n.github.io Udacity’s Machine Learning Nanodegree goo.gl/ODpXj4 Totally new to ML? Recipes goo.gl/KewA03
23. Thank you! Vikram Tiwari @Vikram_Tiwari