Google Cloud Platform Empowers TensorFlow and Machine Learning

Google Cloud Platform Empowers TensorFlow and Machine Learning

• 1.
  Google Cloud PlatformEmpowers TensorFlow and Machine Learning
• 2.
  +Kazunori Sato @kazunori_279 Kaz Sato StaffDeveloper Advocate Tech Lead for Data & Analytics Cloud Platform, Google Inc.
• 3.
  What we’ll cover Whatis Neural Network and Deep Learning Machine Intelligence at Google Scale Cloud Vision API and Speech API TensorFlow and Cloud Machine Learning
• 4.
  What is NeuralNetwork and Deep Learning
• 5.
  Neural Network isa function that can learn
• 6.
  xn > b? w1 wn x2 x1 Mimicking thebehavior of biological neurons
• 7.
  How do you classifythem?
• 8.
  weights bias (threshold) Programmers would specifythe parameters
• 9.
  Let’s see howneural network solves the problem
• 10.
  The computer triesto find the best parameters A neuron classifies a data point into two kinds
• 11.
  Gradient Descent: changingthe params gradually to reduce errors
• 12.
  How do you classifythem?
• 13.
  More neurons =More features to extract
• 14.
  Mapping inputs to afeature space, classifying with a hyperplane From: Neural Networks, Manifolds, and Topology, colah's blog
• 15.
  How about this?
• 16.
  More hidden layers= more hierarchies of features
• 17.
  How about this?
• 18.
  We need tobuild a Deep Neural Network From: Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations, Honglak Lee et al.
• 19.
  From: mNeuron: AMatlab Plugin to Visualize Neurons from Deep Models, Donglai Wei et. al.
• 20.
  Machine Intelligence atGoogle scale
• 21.
  The two bigchallenges of Deep Learning: Computing Power and Training Data
• 23.
  Enterprise Google Cloud is TheDatacenter as a Computer
• 24.
  Jupiter network 10 GbEx 100 K = 1 Pbps Consolidates servers with microsec latency
• 25.
  Borg No VMs, purecontainers 10K - 20K nodes per Cell DC-scale job scheduling CPUs, mem, disks and IO
• 26.
  Confidential & ProprietaryGoogleCloud Platform 26 Google Cloud + Neural Network = Google Brain
• 27.
  What's the scalabilityof Google Brain? "Large Scale Distributed Systems for Training Neural Networks", NIPS 2015 ○ Inception / ImageNet: 40x with 50 GPUs ○ RankBrain: 300x with 500 nodes
• 30.
  Externalizing the power todevelopers
• 31.
  Image analysis withpre-trained models REST API: receives an image and returns a JSON No Machine Learning skill required From $2.50 / 1,000 units (no charge* to try) General Availability Cloud Vision API * You will be charged for Google Cloud Storage and other Google Cloud Platform resources used in your project.
• 33.
  3333 Demo
• 34.
  Pre-trained models. NoML skill required REST API: receives audio and returns texts Supports 80+ languages Streaming or non-streaming Limited Preview - cloud.google.com/speech Cloud Speech API
• 35.
  3535 Demo
• 36.
  TensorFlow and CloudML
• 37.
  Ready to useMachine Learning models Use your own data to train models Cloud Vision API Cloud Speech API Cloud Translate API Cloud Machine Learning Develop - Model - Test Google BigQuery Stay Tuned…. Cloud Storage Cloud Datalab NEW Alpha GA BetaGA Alpha GA GA
• 38.
  Google's open sourcelibrary for machine intelligence tensorflow.org launched in Nov 2015 Used by many production ML projects What is TensorFlow?
• 39.
  # define thenetwork import tensorflow as tf x = tf.placeholder(tf.float32, [None, 784]) W = tf.Variable(tf.zeros([784, 10])) b = tf.Variable(tf.zeros([10])) y = tf.nn.softmax(tf.matmul(x, W) + b) # define a training step y_ = tf.placeholder(tf.float32, [None, 10]) xent = -tf.reduce_sum(y_*tf.log(y)) step = tf.train.GradientDescentOptimizer(0.01).minimize (xent)
• 40.
  TensorBoard: visualization tool
• 41.
  Portable and Scalable Trainingon: Mac/Windows GPU server GPU cluster / Cloud Running on: Android, iOS RasPi
• 42.
  Distributed Training and Predictionwith TensorFlow
• 43.
  Distributed Training withTensorFlow
• 44.
  Data Parallelism split data, sharemodel
• 45.
  ● CPU/GPU scheduling ●Communications ○ Local, RPC, RDMA ○ 32/16/8 bit quantization ● Cost-based optimization ● Fault tolerance Distributed Training with TensorFlow
• 46.
  Tensor Processing Unit ASICfor TensorFlow Designed by Google 10x better perf / watt latency and efficiency bit quantization
• 47.
  TPU on Production RankBrain AlphaGo GooglePhotos Speech and more
• 48.
  Fully managed distributedtraining and prediction Supports custom TensorFlow graphs Integrated with Cloud Dataflow and Cloud Datalab Limited Preview - cloud.google.com/ml Cloud Machine Learning (Cloud ML)
• 49.
  Jeff Dean's keynote:YouTube video Define a custom TensorFlow graph Training at local: 8.3 hours w/ 1 node Training at cloud: 32 min w/ 20 nodes (15x faster) Prediction at cloud at 300 reqs / sec Cloud ML demo
• 50.
  TensorFlow in theWild
• 51.
  TensorFlow powered Fried Chicken Nugget Server From:http://www.rt-net.jp/karaage1/
• 52.
  TensorFlow powered Cucumber Sorter From:http://workpiles.com/2016/02/tensorflow-cnn-cucumber/
• 53.
  TV popstar classifier with95% accuracy From: http://memo.sugyan.com/entry/2016/06/14/220624
• 54.
  Discriminative Localization From: https://github.com/jazzsaxmafia/Weakly_detector
• 55.
  From: http://otoro.net/ Generative Artswith TensorFlow
• 56.
  Thank you!
• 57.
  Links & Resources LargeScale Distributed Systems for Training Neural Networks, Jeff Dean and Oriol Vinals Cloud Vision API: cloud.google.com/vision Cloud Speech API: cloud.google.com/speech TensorFlow: tensorflow.org Cloud Machine Learning: cloud.google.com/ml Cloud Machine Learning: demo video