DATA AND AI APPLICATIONS, TOOLS, TECHNOLOGY DIRECTIONS

Ikhlaq Sidhu, content author
IKHLAQ SIDHU
Chief Scientist and Founding Director
Sutardja Center for Entrepreneurship & Technology
IEOR Emerging Area Professor
Department of Industrial Engineering & Operations Research, UC Berkeley
DATA AND AI
APPLICATIONS, TOOLS, TECHNOLOGY DIRECTIONS

SUTARDJA CENTER
AT BERKELEY
M ETR I C S AT A G LANC E
CHALLENGE
LAB
GLOBAL
PROFESSORS
NZTV
SELF-DRIVING
COLLIDER
DATA-X SCET IN TAIWAN
JOHN BATTELLE
FOUNDER WIRED
MAGAZINE
NEWS
COVERAGE
Undergraduate:
• 12-14 Courses, 1500+
Undergraduates
X at Berkeley: Graduate, Labs and
Professional
• 80+ Grad students
• 100+ Executives
• Labs: Data-X, Blockchain,
Sustainable Food
Ecosystem:
• 14+ Global Partners
• 500+ Executives
• 50+ Investors
MARRISA
MAYER

IEOR135/290AppliedDataScience
withVentureApplications
SupportedbytheData-XLab

• Detection of fake news
• Prediction of long-term energy prices
• Automatic recycling through image recognition
• AI for crime detection, traffic guidance, medical
diagnostics, etc.
• A version of Zillow that is recalculated with the
effects of AirBnB income
• Signal processing and pattern analysis to
improve earthquake warning systems
• Early Autism Detection
• Secure Health Records stored on a Blockchain
find many, many more at:
www.data-x.blog/projects
Data-X Project Examples

• 350 alumni students
• 50% avg enrollment increase / semester
• 80+ great projects completed
• 8+ published research papers
• 100+ industry experts in network
• 20+ students got employment as data
scientists only because they took Data-X
Amazing testimonials:
I think this class is so awesome because it teaches the tools and concepts that are most
commonly used in workplace teams that are involved with data science and applied
machine learning.
We are building on expertise from Data-X, our highly applied
Data Science & AI/ML Lab and Course

Berkeley X-Labs
A New Model for Applied Research Labs
Future of X:
Data, AI,
Blockchain
Expert team
formation
Innovation
Mindset
X
Better University to Industry Connection Models
ü Projects with Business Results
ü Global Teams, Visiting Scholars
ü Company Collaborations
ü Executive Education and Train Trainer Models
ü Self Evaluation Tools
ü Startups Acceleration

Data and AI Application

Basic Concept of Working with Data
• Data Wrangling
• In Production

Real-life Example: ZestCash
• “All data is credit data”
Online Loan Application
Name: JOE SMITH
Name: Joe Smith
The data says: greater credit risk! The data says: lesser credit risk!
Reference: Shomit Ghose
Example: Data and Information is a competitive advantage

Harrah’s Casino: Knowing your customer
• Service provider
of Gambling and
Casinos
• Entry Card
• Pain points
• Intervention
Reference: Supercrunchers

Customer
Insight/
Engagement
Operations:
Reliable &
Predictable
Security &
Fraud
Financial Firms
Network Security
Common Applications as of today …

Who Will Control the Automobile?
Does car need to buy gas? Gas
stations will want to know.
You can sell a new
car once, but you
can sell the data
every minute of
the day
The world’s two largest
sources of data
• Google? or Ford?
– Whoever has the better software and data science
team
– Winner will get the vast (and incredibly valuable)
streams of auto data
Who caused the accident?
Insurance companies will
want to know.
Shomit Ghose

Where Does Data Come From?

Where Does Data Come From?
Real-life Example: ZestCash
• “All data is credit data”
Name: JOE SMITH
Name: Joe Smith
The data says: greater credit risk! The data says: lesser credit risk!
Your Own Web Site Public Data Sets
Stock market, etc.
IOT/Sensors Other Web Sites

Web Scraping
https://github.com/ikhlaqsidhu/data-x/tree/master/03-tools-webscraping-crawling_api_afo
https://github.com/ikhlaqsidhu/data-x

Formatting Data

An ML High Level Framework
• Objects
• Events /
Experiments
• People /
Customers
• Products
• Stocks
• …
In Real Life
Features, but also
loss of information
In Sample
Out of Sample
Person 1
Person 2
Person 3
.
.
.
Person
N
- Characteristics
- Patterns
- Models
- Predictions
- Similarities
- Differences
- Distance
Some data
has observed results

CS: Table
Math: Matrix X, with N rows – each person
m columns, each feature (age, salary, ..)
X =
• Objects
• Events /
Experiments
• People /
Customers
• Products
• Stocks
• …
In Real Life
Features, but also
loss of information
In Sample
Out of Sample
Person 1
Person 2
Person 3
.
.
.
Person
N
- Characteristics
- Patterns
- Models
- Predictions
- Similarities
- Differences
- Distance
Some data
has observed results
An ML High Level Framework

Traditionally 2 Tasks: Classification & Predictive Scoring
The most famous
application has been
recommendation:
“which other user is
most like you”
Extracted Data
often in
Table
Format
Classification:
Cats and Dogs, Speech Recognition
Movie Recommendation
Scoring:
Credit Score, Movie Rating
Heath Score, Any Isoquant…

X Y
X Y
X YML Algorithms Guess
this function F(x)
We have now switched
to Neural Networks as
Function Approximators

Neural net results are close t human results

Data and AI Future Directions

Peter Abbeel – Deep
Reinforcement Learning
Peter Abbeel
Professor at UC Berkeley

Recent AI News
Source: Ken Goldberg, CPAR, People and Robotics Initiative

Does this mean AI Can Do
Everything Better than Humans

Perfect Information vs. Real World
fully observed uncertain
discrete multi-agentsingle agent
infinite time horizon
continuous
finite
Ken Goldberg UC Berkeley
Even then, AI Cannot Solve Real Life Problems Better Than Humans
And in fact, AI Can not even Work without Humans
Ken Goldberg
Leading AI
Researcher at
Berkeley
Professor and
Department Chair,
IEOR
William S. Floyd Jr.
Distinguished Chair

Acknowledgement to Ken Goldberg UC Berkeley
AI Systems Only Work because of Human are Part of the System
Google Operations
People Write Web Pages People at Google Tune
the Results
People Click on What
They Want
Result
Feedback
By clicks
Massive
Data
There is no “Intelligence”, “Desire”, or “Existence” in AI without People
There are only people who “invest in, design and operate the machines”

37
faculty
At Berkeley, we have a lot of research on
“How Machines Will Work as Part of Larger Systems
that Work with People”

My
Drive
Home
From
Berkeley

Autonomous Driving and Driver-Assist
•Communicating intent
•Driver-in-the-loop modeling
•Two-way learning: knowledge
transfer between vehicle and
driver
•Safety in autonomous and
assisted driving
Principal investigators:
Ken Goldberg
UC Berkeley
Anca Dragan
UC Berkeley
Trevor Darrell
UC Berkeley
Francesco Borrelli
UC Berkeley
Ruzena Bajcsy
UC Berkeley
Source: Ken Goldberg, CPAR, People and Robotics Initiative

Ikhlaq Sidhu, content authorSource: Ken Goldberg, CPAR, People and Robotics Initiative
Safety in Human-Robot Interaction:
Guarantees and Verification
Safety-constrained motion
planning for efficiency in factory
human-robot interaction
Learning and prediction for safety
in HRI
Provably safe human-centric
autonomy
Masayoshi Tomizuka
UC Berkeley
Principal investigators:
Claire Tomlin
UC Berkeley
Francesco Borrelli
UC Berkeley

• Large-scale machine learning - amounts of data
• Deep learning - recognition, classification
• Reinforcement learning - time sequence, aided by Neural Networks
• Robotics - beyond navigation, to safe interaction
• Computer vision - most prominent perception, better than human
• Natural Language Processing - interacting with people/dialog
• Collaborative systems - autonomous systems w/people + machines using
complimentary functions
• Crowdsourcing and human computation – harness human intelligence, uses
other AI, vision, ML, NLP, …
• Algorithmic game theory and computational social choice – systems using
social computing, incentives, prediction markets, game theory, peer
prediction, scoring rules, no regret learning
• Internet of Things (IoT) – using AI to unravel sensory information, interfaces,
and protocols
• Neuromorphic Computing – new computing fabrics based on biological
models
http://ai100.stanford.edu/2016-report
New
Data/AI
Systems
Most Common Data/AI Research Trends in 2017

Unsupervised Image to Image
[CycleGAN: Zhu, Park, Isola & Efros, 2017] Pieter Abbeel -- UC Berkeley | Gradescope | Covariant.AI
Generative Models
n “What I cannot create, I do not understand.”
n Ability to generate data that look real entails some form of
understanding
[Radford, Metz & Chintala, ICLR 2016]
Typical CNN converts image to output vector of features

Large Investment and Valuations
Some Very High Valuations
Pieter Abbeel -- UC Berkeley | Gradescope | Covariant.AI
Big Investments in China
Pieter Abbeel -- UC Berkeley | Gradescope | Covariant.AI
Big Investments in China
Pieter Abbeel -- UC Berkeley | Gradescope
Recent Headlines
Pieter Abbeel -- UC B
Recent Headlines
[Macron (France), March 2018] Pieter Abbeel -- UC Berkeley | Gradescope | Covariant.AI

Contact:
Ikhlaq Sidhu
Founding Faculty Director, Center for Entrepreneurship & Technology
IEOR Emerging Area Professor, UC Berkeley
sidhu @berkeley.edu, scet.berkeley.edu

DATA AND AI APPLICATIONS, TOOLS, TECHNOLOGY DIRECTIONS

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Was ist angesagt?

Was ist angesagt? (20)

Ähnlich wie DATA AND AI APPLICATIONS, TOOLS, TECHNOLOGY DIRECTIONS

Ähnlich wie DATA AND AI APPLICATIONS, TOOLS, TECHNOLOGY DIRECTIONS (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

DATA AND AI APPLICATIONS, TOOLS, TECHNOLOGY DIRECTIONS