Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
AI RELOADED - AI APPLICATIONS IN THE INDUSTRYDr. Christoph WindheuserGlobal Head of Intelligent EmpowermentBerlin, November 1st 2018
6000 technologists with 42 offices in 15 countries
Partner for technology driven business transformation
SHAPING THE INDUSTRY AS A THOUGHT LEADER
©ThoughtWorks 2018
AI IS ON THE RISE
“About 30 percent of organizations with more than 1,000 employees have adopted AI, and an additional 25 percent plan to adopt the technologies in the next 12 months.”
Spiceworks Survey 2018(https://www.spiceworks.com/press/releases/spiceworks-survey-tech-spending-expected-rise-budgets-stabilize-2018/)
4
©ThoughtWorks 2018
● Empower and augment human intelligence
● Enable machines to work with humans “hand-in-hand”
● Help humans to become more efficient and work with higher quality
● Automate routine work and give humans more time for difficult and strategic work
● “Forklift for the Brain”
INTELLIGENT EMPOWERMENT
©ThoughtWorks 2018
AGENDA
● It is all about data
● What can AI do today?
● Continuous Intelligence
6
©ThoughtWorks 2018
EMPOWER YOUR DATA
7
● ”Data First” is not just another module – but the redesign of the entire IT Data Architecture
● A “Data Lake” is not the solution!
● Data Management and Data Governance are required
©ThoughtWorks 2018
THE DATA-INFORMATION-INSIGHT-ACTION CYCLE
How data flowsthrough your organization
data
informationinsight
decision
action
©ThoughtWorks 2018
data
acquire
THE DATA-INFORMATION-INSIGHT-ACTION CYCLE
Values attributed to parameters
©ThoughtWorks 2018
data
information
acquire
store, clean, manage, featurize
THE DATA-INFORMATION-INSIGHT-ACTION CYCLE
Values attributed to parameters
Data which has a meaning to a recipient, resolves uncertainty
©ThoughtWorks 2018
data
informationinsight
acquire
store, clean, manage, featurize
model
THE DATA-INFORMATION-INSIGHT-ACTION CYCLE
Values attributed to parameters
Data which has a meaning to a recipient, resolves uncertainty
Understandingand forecasting
©ThoughtWorks 2018
data
informationinsight
decision
acquire
store, clean, manage, featurize
model
productionize
THE DATA-INFORMATION-INSIGHT-ACTION CYCLE
Values attributed to parameters
Data which has a meaning to a recipient, resolves uncertainty
Planning and deciding actions
Understandingand forecasting
©ThoughtWorks 2018
data
informationinsight
decision
actionacquire
store, clean, manage, featurize
model
productionize
execute
THE DATA-INFORMATION-INSIGHT-ACTION CYCLE
Values attributed to parameters
Data which has a meaning to a recipient, resolves uncertainty
Planning and deciding actions
Changing the real world
Understandingand forecasting
©ThoughtWorks 2018
AGENDA
● It is all about data
● What can AI do today?
● Continuous Intelligence
14
©ThoughtWorks 2018
NARROW AI IS COMPOSABLE
15
● Today AI is narrow● ML modules
can be composedlike microservices
● The gradient can be back propagated through the different modules (end-to-end learning)
AI Building Block
Example
Regression:Predict a number
● How much will this house sell for?● How many dresses will sell next week in this shop?
Categorization:Predict a category
● Is this customer NEW, LOYAL, FICKLE, …?● What‘s the probability that this transaction is fraud?● Recommend what a client is most likely to buy next
based on his buying history
Recognize and generate sequences formed by rules
● What is the utterance of this sentence?● What is the sentiment of this sentence?● Translate from language A to B
Learn a strategy of actions to reach a goal(Reinforcement Learning)
● Win Atari Games, Chess, Go● Drive a car● Control an air condition to minimize energy
consumption
Construction: WOOD
EXIF Location
Bedrooms 3
Photo
Price: $1M
Bedrooms:3
Distance to city: 10km
PROJECT EXAMPLE: REAL ESTATE COMPANY (1 / 2)
Construction: WOOD
Location Distance to city: 10km
Condition:GOOD
Photo
Price: $1.1M
Bedrooms:3
PROJECT EXAMPLE: REAL ESTATE COMPANY (2 / 2)
18
PROJECT EXAMPLE: CHAT BOT
Word2Vec
LSTM
©ThoughtWorks 2018
AGENDA
● It is all about data
● What can AI do today?
● Continuous Intelligence
19
©2017 ThoughtWorks Inc. Confidential - please do not distribute.
Wikipedia
Principles of Continuous Delivery(*)
● Create a Repeatable, Reliable Process for Releasing Software● Automate Almost Everything● Build Quality In
(*) From Jez Humble, David Farley: Continuous Delivery, Addison-Wesley, 2011
“Continuous delivery (CD) is a software engineering approach in which teams produce software in short cycles, ensuring that the software can be reliably released at any time.”
CONTINUOUS DELIVERY
©ThoughtWorks 2018
Data Science,Model Building Java Code JAR File
(JVM)
TrainData
Val.Data
Val.Data
ValidationResult 1
ValidationResult 2
IF validation result 1 is equal validation result 2then deploy
equal
IF
Deployment
Java Code JAR File
PROJECT EXAMPLE - PRICE ESTIMATION
22
PROJECT EXAMPLE - CHATBOT
● dvc is git porcelain for storing large files using cloud storage
● dvc connects model training steps to create reproducible workflows
23
master
change-max-depth
try-random-forest
model.pkl
decision_tree.pymodel.pkl.dvc
GCP
dvc - data science version control
24
● We can put data science work into a Continuous Delivery workflow
● Proper data/model versioning tools enable reproducible work to be done in parallel
● Result: Continuous, on-demand AI development and deployment, from research to production, with a single command
● Benefit: production AI systems that are always as smart as your data science team
CONTINUOUS INTELLIGENCE
THANK YOU!Dr. Christoph WindheuserGlobal Head of Intelligent Empowerment