雲端推動的人工智能革命

© 2018, Amazon Web Services, Inc. or its affiliates. All rights reserved.
費良宏
Principal Evangelist @ Amazon Web Services
lianghon@amazon.com
雲端推動的人工智能革命

Quick look at Machine Learning in past year
The MINERvA neutrino detector at Fermi National
Accelerator Laboratory
Visualizing AI detections on the map
Handbag Brand and Color Detection at Condé Nast Amazon Go, sensor-infused store
Spot a Silent Cancer
Self-driving car

“AI Winter Is Well On Its Way” ?
Source: blog.piekniewski.info/2018/05/28/ai-winter-is-well-on-its-way/

A Long History of ML at Amazon
Personalized
recommendations
Inventing entirely
new customer
experiences
Fulfillment
automation and
inventory
management
Drones
Voice driven
interactions

amazon.com, 1995 IEEE internet Computing, 2017

Put machine learning in the hands of every developer
and data scientist
ML @ AWS: Our mission

The Amazon ML Stack
Platform Services
Application Services
Frameworks & Interfaces
Caffe2 CNTK
Apache
MXNet
PyTorch TensorFlow Torch Keras Gluon
AWS Deep Learning AMIs
Amazon SageMaker AWS DeepLens
Rekognition Transcribe Translate Polly Comprehend Lex

Platforms & Frameworks

Bottom Layer: Frameworks & Interfaces
NVIDIA
Tesla V100 GPUs
P3
AWS Deep Learning AMI
5,120 tensor cores
128 GB of memory
1 petaflop of compute
NVLink 2.0
~14X faster than P2

Deep Learning Frameworks Outlook
TensorFlow
• Dynamic computational graph (eager execution)
• Publish & discovery model (TensorFlow Hub)
• More languages (TensorFlow.js, TensorFlow for Swift)
• Lightweight, cross-device solution (TensorFlow Lite)
• More hardware support (Integration with TensorRT, MKL-DNN and TPU)
• Bayesian analysis (TensorFlow Probability)
…
Apache MXNet/Gluon
• MXNet 1.2.0 released
• Adds support for Keras 2
• Deep learning for NLP (GluonNLP)
• Computer vision Toolkit (Gluon CV Toolkit)
• The book of ”Deep Learning - The Straight Dope” is coming soon
…
PyTorch
• Caffe2 merges with Pytorch
• PyTorch 1.0 , the milestone release
• exporting to C++-only runtimes
• optimizing mobile systems on iPhone, Android, Qualcomm etc.
• quantized inference (such as 8-bit inference)
• JIT-Compiler for models (torch.jit)
• Not making any big changes to the existing API

AWS Deep Learning AMIs
Support for deep learning frameworks Accelerate model training
Conda AMI
For developers who want pre-
installed pip packages of deep
learning frameworks in separate
virtual environments. Available in
in Ubuntu and Amazon Linux and
Windows 2016 versions.
Base AMI
For developers who want a clean
slate to set up private deep
learning engine repositories or
custom builds of deep learning
engines. Available
in Ubuntu and Amazon
Linux versions.
AMI with source code
For developers who want
preinstalled deep learning
frameworks and their source code
in a shared Python environment,
available for P3 instances in CUDA
9 Ubuntu and Amazon
Linux versions.

ML is still too complicated for everyday developers
Collect and prepare
training data
Choose and
optimize your ML
algorithm
Set up and manage
environments for
training
Train and tune
model
(trial and error)
Deploy model
in production
Scale and manage
the production
environment

Amazon SageMaker
Collect and prepare
training data
Choose and
optimize your ML
algorithm
Set up and manage
environments for
training
Train and tune
model
(trial and error)
Deploy model
in production
Scale and manage
the production
environment
Easily build, train, and deploy machine learning models

Amazon SageMaker
Pre-built
notebooks for
common
problems
K-Means Clustering
Principal Component Analysis
Neural Topic Modelling
Factorization Machines
Linear Learner - Regression
XGBoost
Latent Dirichlet Allocation
Image Classification
Seq2Seq
Linear Learner - Classification
ALGORITHMS
Apache MXNet
TensorFlow
Caffe2, CNTK,
PyTorch, Torch
FRAMEWORKS
Set up and manage
environments for
training
Train and tune
model (trial and
error)
Deploy model
in production
Scale and manage the
production environment
Built-in, high-
performance
algorithms
Build

Amazon SageMaker
Pre-built
notebooks for
common
problems
Built-in, high-
performance
algorithms
One-click
training
Hyperparameter
optimization
Build Train
Deploy model
in production
Scale and manage
the production
environment

Amazon SageMaker
Fully managed
hosting with auto-
scaling
One-click
deployment
Pre-built
notebooks for
common
problems
Built-in, high-
performance
algorithms
One-click
training
Hyperparameter
optimization
Build Train Deploy

Model hosting (SM)
Real-time Fraud Detection in AWS with Amazon
SageMaker
Calculate
features
Reader
Cleanser
Processor
Data
Look-up
Training
Feature store
Model training (SM)
Model
Client service

AWS DeepLens
HD video camera
Custom-designed
deep learning
inference engine
Micro-SD
Mini-HDMI
USB
USB
Reset
Audio out
Power
HD video camera
with on-board
compute optimized
for deep learning
Tutorials, examples,
demos, and pre-
built models
From unboxing
to first inference
in <10 minutes
Integrates with
Amazon SageMaker
and AWS Lambda
10
MIN
The world’s first deep learning-enabled video camera for developers

Get Started with Deep Learning
B u i l d c u s t o m d e e p - l e a r n i n g m o d e l s i n t h e c l o u d u s i n g
A m a z o n S a g e M a k e r , o r u s e t h e c o l l e c t i o n o f p r e t r a i n e d
m o d e l s i n c l u d e d w i t h A W S D e e p L e n s
It takes less than 10 minutes with AWS DeepLens

AWS Machine Learning
Languages Services

Powerful language capabilities
Amazon Transcribe
Automatic conversion of speech into accurate, grammatically correct
text
Amazon Translate Natural and fluent language translation
Amazon Polly Turn text into lifelike speech using deep learning
Amazon Comprehend Discover insights and relationships in text
Amazon Lex Conversational interfaces for text-based and voice-based applications

Benefits of ML Language Services
• Easily add intelligence to apps
with pre-trained APIs for speech, transcription, translation, language analysis, and
chatbot functionality
• Connect to comprehensive analytics
including data warehousing, business intelligence, batch processing, stream processing,
and workflow orchestration
• Integrate with the most complete big data platform
including the data lake and database tools to run machine learning workloads

Common Language Use Cases
Information Bots
Education
Accessibility
Knowledge Management
Voice of Customer
Applications
Customer Service/
Call Centers
Enterprise
Digital Assistant
Semantic Search
Captioning Workflows
LocalizationPersonalization

Amazon Comprehend
Discover insights and relationships in text using deep learning
Entities
Key Phrases
Language
Sentiment
Amazon
Comprehend

Amazon Comprehend
Storm
World Series
Stock market
Washington
Library of news
articles
Amazon
Comprehend
Discover insights and relationships in text using deep learning

Amazon Transcribe
Automatic conversion of speech into accurate, grammatically
correct text
Support for
telephony
audio
Timestamp
generation
Intelligent
punctuation and
formatting
Recognize
multiple
speakers
Custom
vocabulary
Multiple
languages

Amazon Translate
Natural and fluent language translation
Real-time
translation
Batch analysis Automatic language
recognition
Low cost

Amazon Polly
Turn text into lifelike speech using deep learning
Wide selection of
voices and languages
Synchronize
speech
Fine-grained
control
Unlimited replay

Amazon Lex
Building conversational interfaces into any application using
voice and text
Integrated
development in
the AWS console
Trigger
Lambda
functions
Multi-step
conversations
One-click
deployment
Enterprise
connectors
Fully
managed

AWS Lambda
Amazon S3
Amazon
Athena
Integrating Speech, Translation, and NLP + AWS
Audio Input
Amazon
QuickSight
Amazon
Comprehend
Amazon
Transcribe
Amazon
Translate

Lex Use Case: Digital Assistant to Book a Hotel
Book hotel
Taipei
“Book a hotel in
Taipei”
Automatic speech
recognition
Hotel booking
Taipei
Natural language
understanding
Intent/slot
model
UtterancesHotel booking
City Taipei
Check in November 30
Check out December 2
“Your hotel is booked for
November 30.”
Amazon Polly
Confirmation: “Your hotel is
booked for November 30.”
“Can I go ahead
with the booking?”
a
in

“Building intelligent applications to help customers drive their businesses is our entire focus. Amazon Comprehend allows us to
analyze unstructured text within search, chat, and documents to understand intent and sentiment. This capability enables us
to train our Coleman AI skillset, and also provide a truly focused and tailored search experience for our customers.”
Manjunath Ganimasty, V.P. Software Development with Infor
“Through Amazon Lex, we're adding sophisticated natural language processing capabilities that help GrowthBot provide a more
intuitive UI for our users. Amazon Lex lets us take advantage of advanced AI and machine learning without having to code the
algorithms ourselves”
Dharmesh Shah, HubSpot CTO and Founder
“At Isentia, we built our media intelligence software in a single language. Having tried multiple Machine Translation services in the
past, we are impressed with how easy it is to integrate Amazon Translate into our pipeline and its ability to scale to handle any
volume we throw at it. The translations also came out more accurate and nuanced and met our high standards for clients.”
Andrea Walsh - CIO, Isentia
AWS ML Language Services Customers
“I can't think of many use cases where accurate pronunciation is more important than when you're learning a new language. We have
found that the Amazon Polly voices are not just high in quality, but are as good as natural human speech for teaching a language.”
Severin Hacker, CTO, Duolingo
“RingDNA is an end-to-end communications platform for sales teams. A critical component of RingDNA’s Conversation AI requires
best of breed speech-to-text to deliver transcriptions of every phone call. RingDNA is excited about Amazon Transcribe since it
provides high-quality speech recognition at scale, helping us to better transcribe every call to text.”
Howard Brown – CEO & Founder, RingDNA

AWS Machine Learning
Computer Vision Services

Amazon Rekognition
Fully
managed
service
Easy-to-use
API
Low costImage
analysis
Video
analysis
Easy-to-use deep learning-based computer vision analysis

Amazon Rekognition Image
Object and Scene
Detection
Facial
Analysis
Face
Recognition
Text in Image
Unsafe Image
Detection
Celebrity
Recognition
Detect objects, scenes, and faces, extract text, recognize celebrities, and identify unsafe content in images
Face
Comparison

Video Analysis
Limitations of traditional solutions
Temporal information lost Motion context lost

Amazon Rekognition Video
Media and entertainment
Create metadata for celebrities, emotions, key topics in video
with time segments for recommendation engines and ad
placement
Automatically detect unsafe
content, based on market requirements
Extract data in streaming mode to enhance
user engagement

Use case: Video search index
Video Amazon S3 AWS Lambda Amazon Rekognition Video
Amazon Elasticsearch Amazon DynamoDB
1. Video is uploaded
and stored to S3
2. Amazon Rekognition Video
creates metadata for celebrities,
emotions, key topics in video with
time segments for search
4. Lambda also pushes
the metadata and confidence
scores into Elasticsearch
3. The output is persisted as
metadata into DynamoDB
to ensure durability

Public safety
Recognize a person of interest across a collection
of millions in real time across hundreds of cameras
Track a person of interest across a video
Create alerts by detecting objects and activities
of interest, such as car, license plate, running, etc.

Use case: Public safety immediate response
Live Street Camera Amazon Kinesis Video Streams Amazon Rekognition Video Face collection
1. Camera-captured video
streams are processed by
Kinesis Video Streams
2. Amazon Rekognition Video analyses
the video and searches faces on screen
against a collection of millions of faces
User
3. User is notified
in case of face matches
Amazon SNS AWS Lambda Amazon Kinesis
Streams

Face Search — Public Safety

Face Search—Media and Entertainment
A u t o m a t i n g F o o t a g e
T a g g i n g w i t h A m a z o n
R e k o g n i t i o n
Indexed 99,000 people
Saves ~9,000 hours a year in labor

Amazon Rekognition Customers

Why AI ? Why AWS ?

Customers running machine learning on AWS today

Artificial
Intelligence
Object store
Database
Datawarehouse
Stream analysis
Business Intelligence
Hadoop
Spark/Presto
Elasticsearch
AI-Driven “Flywheel” model
Click stream
User activity
UGC
Purchas
click
Preference
Sensor data
More
data
Better
Analytics
Better
Product
More
users

Algorithm
DataCloud
GPUs
& Computation resource
Today is a new AI age

The trends AI will nourish
1. AI will proliferate natural, contextual human-computing
interfaces.
2. AI will power smart IoT and fluid application integration.
3. Computing ecosystems will interconnect AI-enabled software.
"The enterprise that does not innovate inevitably ages
and declines.
And in a period of rapid change such as the present ... the
decline will be fast."
— Peter F. Drucker

謝謝!

雲端推動的人工智能革命

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Was ist angesagt?

Was ist angesagt? (20)

Ähnlich wie 雲端推動的人工智能革命

Ähnlich wie 雲端推動的人工智能革命 (20)

Mehr von Amazon Web Services

Mehr von Amazon Web Services (20)

雲端推動的人工智能革命