48
Machine Intelligence made easy: Vision/Speech API, TensorFlow and Cloud ML

Machine Intelligence made easy: Vision/Speech API, TensorFlow

Embed Size (px)

Citation preview

Page 1: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Machine Intelligence made easy:

Vision/Speech API, TensorFlow and Cloud ML

Page 2: Machine Intelligence made easy: Vision/Speech API, TensorFlow

+Kazunori Sato@kazunori_279

Kaz Sato

Staff Developer AdvocateTech Lead for Data & AnalyticsCloud Platform, Google Inc.

Page 3: Machine Intelligence made easy: Vision/Speech API, TensorFlow

What we’ll cover

What is Neural Network and Deep Learning?

Machine Intelligence at Google Scale

Cloud Vision API and Speech API

TensorFlow and Cloud Machine Learning

Page 4: Machine Intelligence made easy: Vision/Speech API, TensorFlow

What is Neural Network and Deep Learning?

Page 5: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Neural Network is a function that can learn

Page 6: Machine Intelligence made easy: Vision/Speech API, TensorFlow

0.88 (cat)

0.12 (dog)

0.01 (car)

input vector (pixel data)

output vector (probability)

Mimics neurons with matrix operations

Page 7: Machine Intelligence made easy: Vision/Speech API, TensorFlow

How do you classify them?

Page 8: Machine Intelligence made easy: Vision/Speech API, TensorFlow

The computer tries to find the best parameters

Let’s try with neural network:

Page 9: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Computer tries moving the paramsgradually to reduce errors

Page 10: Machine Intelligence made easy: Vision/Speech API, TensorFlow

How do you classify them?

Page 11: Machine Intelligence made easy: Vision/Speech API, TensorFlow

More hidden layers x neurons = More complex patterns

Page 12: Machine Intelligence made easy: Vision/Speech API, TensorFlow

How about this?

Page 13: Machine Intelligence made easy: Vision/Speech API, TensorFlow
Page 14: Machine Intelligence made easy: Vision/Speech API, TensorFlow

How about this?

Page 15: Machine Intelligence made easy: Vision/Speech API, TensorFlow

0.00 (0)0.00 (1)0.00 (2)0.00 (3)0.00 (4)0.00 (5)0.00 (6)0.00 (7)1.00 (8)0.00 (9)Even a single layer

can yield about 90% accuracy

Page 16: Machine Intelligence made easy: Vision/Speech API, TensorFlow

How about this?

Page 17: Machine Intelligence made easy: Vision/Speech API, TensorFlow

From: Convolutional Deep Belief Networks for Scalable Unsupervised Learning of Hierarchical Representations, Honglak Lee et al.

We need many more s

Page 18: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Machine Intelligence at Google scale

Page 19: Machine Intelligence made easy: Vision/Speech API, TensorFlow

The two big challenges of Deep Learning:Computing Power and Training Data

Page 20: Machine Intelligence made easy: Vision/Speech API, TensorFlow
Page 21: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Enterprise

Google Cloud is

The Datacenter as a Computer

Page 22: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Jupiter network

10 GbE x 100 K = 1 Pbps

Consolidates servers with

microsec latency

Page 23: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Borg

No VMs, pure containers

10K - 20K nodes per Cell

DC-scale job scheduling

CPUs, mem, disks and IO

Page 24: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Confidential & ProprietaryGoogle Cloud Platform 24

Google Cloud +

Neural Network =

Google Brain

Page 25: Machine Intelligence made easy: Vision/Speech API, TensorFlow

What's the scalability of Google Brain?

"Large Scale Distributed Systems for Training Neural

Networks", NIPS 2015

○ Inception / ImageNet: 40x with 50 GPUs

○ RankBrain: 300x with 500 nodes

Page 26: Machine Intelligence made easy: Vision/Speech API, TensorFlow
Page 27: Machine Intelligence made easy: Vision/Speech API, TensorFlow
Page 28: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Externalizing the power of Brainto developers

Page 29: Machine Intelligence made easy: Vision/Speech API, TensorFlow
Page 30: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Image analysis with pre-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.

Page 31: Machine Intelligence made easy: Vision/Speech API, TensorFlow
Page 32: Machine Intelligence made easy: Vision/Speech API, TensorFlow

3232

Demo

Page 33: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Pre-trained models. No ML 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

Page 34: Machine Intelligence made easy: Vision/Speech API, TensorFlow

3434

Demo

Page 35: Machine Intelligence made easy: Vision/Speech API, TensorFlow

TensorFlow

Page 36: Machine Intelligence made easy: Vision/Speech API, TensorFlow

The Machine Learning Spectrum

TensorFlow Cloud Machine Learning Machine Learning APIs

Industry / applications

Academic / research

Page 37: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Google's open source library for

machine intelligence

tensorflow.org launched in Nov 2015

The second generation

Used by many production ML projects

What is TensorFlow?

Page 38: Machine Intelligence made easy: Vision/Speech API, TensorFlow

# define the networkimport tensorflow as tfx = 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 stepy_ = tf.placeholder(tf.float32, [None, 10])xent = -tf.reduce_sum(y_*tf.log(y))step = tf.train.GradientDescentOptimizer(0.01).minimize(xent)

Page 39: Machine Intelligence made easy: Vision/Speech API, TensorFlow

TensorBoard: visualization tool

Page 40: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Portable● Training on:

○ Data Center

○ CPUs, GPUs and etc

● Running on:

○ Mobile phones

○ IoT devices

Page 41: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Tensor Processing Unit

ASIC for TensorFlow

Designed by Google

10x better perf / watt

latency and efficiency

bit quantization

Page 42: Machine Intelligence made easy: Vision/Speech API, TensorFlow

RankBrain

AlphaGo

Google Photos

Speech

Page 43: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Fully managed, distributed training and prediction

for custom TensorFlow graph

Supports Regression and Classification initially

Integrated with Cloud Dataflow and Cloud Datalab

Limited Preview - cloud.google.com/ml

Cloud Machine Learning (Cloud ML)

Page 44: Machine Intelligence made easy: Vision/Speech API, TensorFlow

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

Page 45: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Summary

Page 46: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Ready to use Machine 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

AlphaGA

GA

Page 47: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Links & Resources

Large Scale 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

Page 48: Machine Intelligence made easy: Vision/Speech API, TensorFlow

Thank you!