"LLMs for Python Engineers: Advanced Data Analysis and Semantic Kernel",Oleks...
The AI Spring - Innovations for the next decade
1. The AI spring
- Innovations for the next decade
May 2015
1
Pari Natarajan, Hardik Tiwari
Rishabh Saraswat, Amrit Murali
2. Introduction of
Turing test
AI has evolved rapidly in the last few years
1950
Computational Power
Data Platforms
Better Algorithms
First AI program
to play Tic Tac Toe
1960
RDBMS
LOGIC THEOREMS
-Single layer learning,
Perceptron, Adaline
IBM Deep blue
defeats Gary Kasparov
1997
OLAP
NEURAL NETWORKS
- Multilayer Back propagation
512
Core GPU
2880
Core GPU
12000
Core GPU
Cost of Computing
$200
Per million transistors
$50
Per million transistors
$0.05
Per million transistors
2011 2015
Watson became
Jeopardy Champion
BIGDATA PLATFORMS
-HDFS
DEEP LEARNING
-Convoluted Neural Network
2
DeepMind’s self-taught AI can
beat human players at 29 of 49
Atari games
3. • 10360 possible moves
• Monte Carlo tree search & Q Learning
• Statistical, learned and general purpose
• Learned from 30 million moves
• 1023 Trillion Possible Outcome
• Brute Force Algorithm
• Symbolic, hand crafted and domain specific
• 700,000 Grandmaster chess games
Deep Blue AlphaGo
And is able to beat human champions in complex board games..
“That’s a strange move. I thought it was a mistake.”
– Lee Sedol
It may be a hundred years before a computer beats humans at Go —
maybe even longer!!
-AI Experts in 1997, NY times
3
4. “You get in the car, press the button, and
it will drive you all the way from
Mountain View to San Francisco”
- George Hotz
2015
Built Autonomous Car in a month in his
garage
<$1000
Off-the-shelf ADAS Kit in Beta
$3.1 Million
Funding received
3
Total Team Size
7.25
Hours worth of driving data open sourced
AI has moved on from games to the real world
4Source: Comma.ai
5. MANUFACTURING FINANCIING
OWNERSHIP
EXPERIENCE
DESIGN &
DEVELOPMENT
AI-Designed Car
Developing the first AI
designed car
Partnered with the
robotics and intelligent
systems group to drive
innovations in cognitive
systems in factories
Intelligent Systems
PRODUCT
FEATURES
Set up a $25 million
research centre in
collaboration with MIT for
autonomous vehicle
technologies
Autonomous Vehicles
AI Based Designs
Simulated Testing
Baidu invested in
ZestFinance, a startup
that uses machine
learning to develop a
credit scoring platform.
Credit Scoring
ADAS
Connected Car
Speech Recognition
Intelligent Production
Line
Integrated Systems
Ride Sharing
On-Demand Transport
Credit Scoring
Fraud Detection
Predictive Modelling
Tesla’s Autopilot can, in
real time, learn the daily
routes taken by it’s users.
Ride-Sharing
$721 MnTotal Funding
2011Average Founding year
192Disruptors
Adds value across the industry value chain
Automotive
5Source: GEIP
6. And is disrupting all industry verticals
1- Analysed basis data maturity, software penetration, regulatory restrictions across the value chain representing disruption potential over next 5 years
2 – Analysed basis current investments ( talent + acquisition) for all players
NLP platform
Alexa
BFSI
Healthcare
PotentialtoDisruption1
AI Maturity 2
Retail
Predictive diabetes
management
AI based Robo advisory service.
Enterprise Software
Semicon
Consumer Electronics Microsoft Cortana and
Intelligent Cloud
Machine Learning Enabled
Hardware
NASA software to enable damaged aircrafts, find a safe
landing spot.
Recommendation based on
photographs
Autonomous driving
Auto
Aerospace
Consumer Software
Machine
Learning enabled
Advertising.
6Source: GEIP
7. Artificial Intelligence is a sum of many algorithms..
Constraint Satisfaction
Probabilistic reasoning
Logical Reasoning
Machine Learning
Search & optimization
Control theory
Facial Recognition Autonomous VehiclesDrones & Robotics Personalized
Marketing
Financial Trading
Artificial Intelligence Building computer programs that make data driven decisions without being explicitly
programmed and also automatically improve over time
What is
NLP
Computer Vision
7
8. Deep Learning is a branch of Neural Networks, where there are ‘n’ intermediate processing layers between the input and
output.
WHY DEEP LEARNING
Automated Feature Extraction – Can recognize patterns by itself
Less training data required
Reuse the same algorithm or different applications
Deep Learning Algorithms
Deep Boltzmann
Machine
Deep Belief
Network
Convoluted
Neural Networks
Stacked Auto
Encoders
Input
Input
Input
Output
Convoluted Neural Nets
‘ n’ Intermediate Layers
Output
Output
Input
Input
NLP
Natural Language DB Queries Spam Detection
Early Detection of curable diseases
Personalized Ad Recommendations
Automated Conversation Coach
DAVE2 – Autonomous Vehicle
Vision
Pattern Recognition
Speech Recognition
Speech to Text Conversion
SIRI
Deep Search 2 for voice search
Deep learning is the most dominant of all
8
9. AI innovations are dominated by Tech Mafias and Start-ups
Hardware
APPLICATIONS
PLATFORMS
INFRASTRUCTURE
MACHINE INTELLIGENCE NLP COMPUTER VISION
DEEP LEARNING
ADAS
GESTURE CONTROL
Enterprise Software Assistants
Productivity
FinTech
AdTech
Data Platforms
HealthTech
Auto
Dominated by the startups who build
verticalized applications for various
use cases.
Applications – The Startup Zone
Focused their efforts on building
platforms that can then be leveraged
by the ecosystem.
Platforms – Tech Mafia
Playground
The AI focused companies can be
found providing the infrastructure that
enables the rest of the landscape.
Infrastructure – G500
Domination
Intensity
Start-ups G500Tech Mafia
9Source: GEIP
10. Quarterly funding trend (2013-16 YTD)
Q1, 2012 Q1, 2013 Q1, 2014 Q1, 2015 Q1, 2016
$94
$137
$253
$121
$302
$552
$926
$901
$602
$1,049
Raises $100M for Deep
learning based ultrasound
Google acquires Deepmind for
$500M
Raises $65M for ML based
threat
detection
Q1, 2011
Focussing on reverse
engineering the neocortex
raised series A
Ten fold increase in start-up funding in the last five years
200
USD
0.4B
APPLICATIONS
600
USD
2.5 B
PLATFORMSENABLERS
1400
USD
11 B USD
1 B
USD
0.9 B
USD
0.7 B
Healthcare
Fintech
Auto
USD
2.3 B
USD
0.7 B
USD
0.5 B
USD
0.3 B
USD
0.1 B
Machine
Learning NLP Computer
Vision
Data Platforms
Hardware
USD
6 B
Enterprise
Software
#Startups Funding 10Source: GEIP
11. 2277
Start-ups $14.28 Billion
Global AI start-up distribution
USA
China
Germany
Spain
UK
Israel
85
40
55
188
1170
55
169
India
86
Canada
Netherlands
25
26
Australia
Brazil
18
France
43
25 Singapore
Hong Kong12
Total FundingNumber of Start-ups
And is dominated by the US
Global AI start-up distribution
$11.5 B
$0.6B
$0.3 B
$0.5 B
$0.6 B
$0.1 B
$0.1 B
$0.1 B
APPLICATIONS-FINTECH, HEALTHCARE
APPLICATIONS –HRTECH, HEALTHCAREAPPLICATIONS – AUTO, FINTECH, RETAIL
PLATFORMS- DEEPLEARNING, VISION
ENABLERS –BIGDATA PLATFORMS
Computer
Vision
AI based consumer
robotics start-up
Massively scaled
deep learning
Threat
detection
ML based
recruitment
ML for retail
Personalised
healthcare
NLP API
PLATFORMS- DEEPLEARNING, NLP
11Source: GEIP
12. European start-up ecosystem is evolving and leveraged by US companies
441
Start-ups
$ 1 B
Investment
APPLICATIONS
PLATFORMS
INFRASTRUCTURE
HARDWARE SOFTWARE
DEEP LEARNING- $
170M
NLP - $33M
SMART ROBOTS -$20M
AUTO - $64M FINTECH - $97M
HEALTHCARE -$115M
ADTECH - $26M
SECURITY -$100MPRODUCTIVITY -
$130M
COMPUTER VISION - $50M
USGlobalEU
60%30%10%
30
Acquisitions
$1.2 B
Investment
Start-ups
Deep Learning
NLP
Robotics
Computer Vision
Productivity
12Source: GEIP
13. Healthcare
Home Automation
CloudPersonal Assistants
Tech Mafias are building an AI-first future
Automotive Wearables
AI Talent Acquisition Patents -30K $10 B 300+
Wearables for
Health
monitoring
Project
Titan
Siri controlled
home kit
Apple
Smartwatch
iPhone iOS 10 image
recognition
Spotlight for images
& text
Siri
Aerospace
Google
Home
Verily –algorithms for
Diagnosis Deepmind for
Healthcare
Google X
nanoparticle
research
Google X –Self
Driving
Android Wear
Smartwatch
Google
Loon
AI Robot-
GoogleX
Google Prediction
API
AlloGoogle Now
Tensorflow
Project Jacquard
Deepmind Google for Work
AI Platforms
Cortana for
Healthcare
Microsoft –Volvo
Self driving
Kinect
Cortana
SwiftKey
Microsoft Graph –
Sales lead scoring
Hololens
Azure ML
DSSTNE
Oculus
UAV Aquila
Facial
recognition
Facebook M
Facebook
Deeptext
FAIR
Wit.ai
AlexaDrone
Delivery
Recommender
systems
AWS ML
CNTK
13Source: GEIP
14. And are acquiring younger start-ups to accelerate innovation pace
Amazon Apple Facebook MicrosoftGoogle
Average Acquisition Year
AcquireeMaturity
Dot Com Era Smartphone Era Cognitive Era
2004 2010 2016
1
2
3
4
5 Bulk of the acquisitions by Microsoft and
Google to boost their Search Tech.
MS and Apple begin work
on Gesture Control devices.
The Tech Mafia investing heavily in
AI enablement platforms
Google begin
work on Maps
NLPVision
ML NLP Vision ML NLP Vision Robotics
14
Indicates Average
15. Leveraging global talent pool
9300
19600
Seattle Area
Bay Area
2100
3200
Boston
New York
9502700
Bangalore
1600
• Traditional Hubs for Engineering
for the Tech Mafia - Machine
learning to NLP & Computer
vision.
• Driverless Cars, Drones, Data
Science, Cyber Security are the hot
areas
West Coast of USA East Coast of USA Western Europe & IsraelIndian Hi-Tech
Cities
460
Singapore
660
BeijingIsrael
Hyderabad
• Top universities like CMU &
MIT have hot focus on Artificial
Intelligence research;
• EU’s Human Brain Project
for spending close to 1 billion
euros on AI over the next
decade.
• OEM’s like Renault,
Volkswagen partnering with
Autonomous start-ups like
Mobileye
• IBM set up its Watson
unit in India in 2012 to
work for Healthcare and
BFSI clients in US.
• Baidu is investing in deep
speech for voice-based
searches that leverage
speech recognition;
910
Spain
4100
UK
2000
France
3300
Germany
Hong Kong & Singapore
X ER&D Workforce in AI
950
Netherlands
TECH MAFIA HOTSPOT UNIVERSITY RESEARCH AUTO OEMs AI FOCUS OVERSEAS FOCUS CHINESE INTERNET AI DRIVERS
45%
55%
Tech Mafias
Rest
Tech Mafias own 45%
of global AI talent
15Source: GEIP
16. X
User Base on
GITHUB
And opening their innovation to others to build on
Computer Vision
KINDRED
Robotics
Project Malmo
Tensorflow
DSSTNE
35K
Facebook
for Torch Swift-AI
CNTK
DSSTNE is designed to support
problems with sparse data. 3KAI research built on top of the
game Minecraft.2K
Significantly faster than the default
Torch and allow users to train larger
neural nets
A unified deep-learning toolkit
that describes neural networks
as a series of computational
steps
6K
Swift AI is a high-performance AI and
machine learning library
Open sourcing their APIs
allows the Tech Mafia to
democratize innovation.
1K 1K
Computer Vision for
refrigerators
Most popular Open Source AI Library.
User base has grown tenfold since it’s
release in Nov. 2015.
Makoto Koike uses TensorFlow to
sort Cucumbers
Cornel University project
on Cyber-Security
Projects based on Platforms
16
17. “Success in creating AI would be the biggest event in human history. It might also be the last, unless we learn how to avoid the risks.”
- Stephen Hawking
Public failures not withstanding..
Trending fake news articles 7 reported accidents (1 fatal) since April
2016
Facebook Tesla
Google
Deepmind failed at describing dumbbells Microsoft’s Tay became a racist bot
Microsoft
17
18. Top R&D spenders are lagging behind..
Plan to release Xeon Phi processor
line for AI applications-~$400M
investments
Leveraging ML for network threat
products – Cognitive Threat
Analytics
Invested $5B in building an AI
powered Gigafactory.
$1B investment to establish the
Toyota Research Institute for AI
$500M investment in a 200
member AI R&D lab in Sillicon
Valley
1 : Investment in AI in terms of talent & acquisition or Funding raised ( for startup)
2 : Focus on emerging technologies vs Older algorithms, Focus on Ecosystem creation and Platform adoption/maturity
Google
Microsoft
Amazon
Facebook
Apple
IBM
Bosch
Volkswagen
Intel
Oracle
Cisco
Foxconn
SAP
Airbus
Mobileye
Sentient
Zoox
Datarobot
X.ai
AI Focus1
FutureReadiness2
Startups G500Tech Mafia
G500
Start-ups
Tech Mafia
Size of the bubble indicates R&D spend
Source: GEIP 18
19. But can accelerate through four simple steps
Identify Business Case
1
Build a Data Ecosystem
2
Collaborate with the Ecosystem
Leverage newer talent hotspots
4
3
Algorithm &
Platforms
Data Talent
Use Cases
AI Puzzle
19
20. Identify and prioritize AI’s role
DATA
COMPLEXITY
APPLICATION COMPLEXITY
Wide range of interconnected activities
Well-defined rules, procedures and criteria
Complete Autonomy
Augment Humans
Reliant on individual expertise and experience
Original, innovative work
Surgical RobotsChatbots
Echo, the home control device
Automated meetings scheduler
Image Search
Auto- Recommendations
Automated Factories
Robo- Advisory
Enterprise security through AI
Autonomous Car
DeepLearningRuleBasedEngine
AI bot designed car
AI based website design platform
20
21. Build a Data ecosystem and own it
Conceptualization
Design &
Development
Product Usage ServiceabilityManufacturing
ERP
Geo-location
data
Social
data
Compete
Data
Usage
data
Bug Reports
Product
Data
Customer
Data
Open Data
Web Data
Partner
Data Market Data
Design Data
Libraries
Enterprise Data
Customer
Map
Usage
data
Govt.
Data Content
Logs
SCADA
Sensor
Market Data
Machine Data
Energy Pricing MRO data
PLM
Product
Cloud
Supplier
APIs
Data as a Product External Data APIs Internal Data APIs
21
22. Toyota leverages multiple facets of the innovation
fabric to drive innovations in aligned technology
areas
Headquarters in Japan is supplemented
by 5 engineering hubs
India, China, Thailand ,Mexico & Brazil
Palo Alto research labs for AI & Robotics
research
Acqui-Hired IT born robotics start-up
The SAIL-Toyota Centre for AI Research
Partnered to develop autonomous car
technology.
New Age Innovation
Fabric
Porous innovation permeating beyond
the walls of the organization
Balance internal and external innovation
22
23. Venture Fund
Accelerators
Evangelize
Ecosystem
Collaboration
• Extensive hands on support &
infrastructural support
• Connecting with clients and investors
• Limited platform & soft infrastructure support
• Connect the startup teams with VCs & partners
• Partnership with accelerators/Universities
• Mentorship support & events participation
Capital Investment Non-Capital Investment Examples
• Product GTM support
• Senior level team hiring /restructuring
Partner with start-ups
3rd Party
Accelerators
• Partner with other accelerators,
innovation workshops with stakeholders
Arena 120
Microsoft Start-up
accelerator
23Source: GEIP
24. Build the right platform partnerships
GPU Hardware Platforms
Big Sur Tensor Processing Unit
Github – Public
Datasets
Infrastructure Platforms
Data Sets
Applications
ORCHESTRATION OF SPECIALISTS
Leverages Watson’s open API
to build MyCareLink Smart App
that predicts low blood sugar
Apache 2.0 open
source libraries
EHR, Clinical data
through pharmacies
and universities
H20.ai python based ML
libraries
AWS ML optimized
infrastructure
Fraud Detection Diabetes Detection Cart checkouts Drones
AI Platforms
Driver Safety
24
EXAMPLES
25. Leverage global talent
USA
Canada
65 110
10
UK
35
China
70
85
Available AI & related Talent in Country (‘000)
2015 2025
20
Israel
India
A global satellite and is a part of the global CoE for
ML reporting to Berlin
Global engineering hub and has small team
working on ML
Global engineering satellite and drives activities in
Computer vision space
Global CoE for IoT and Advanced ML space.
18
6
15 155
Beijing a global engineering satellite and works on
NLP and Computer vision tech
Partnered with Didi for crowdsourcing
challenge for optimising route algorithm
25Source: GEIP
26. Some of the European peers have a head start.
Enterprise Software
Have opened up multiple research labs all over the
world, focused on collaborating with start-ups and
scientific institutions with 420M investment.
Intelligent Machines
BMW is teaming up with Mobileye and Intel to
work on its first fully autonomous self-driving
car, which it plans to ship by 2021
Self-Driving Car
Smart Car
Audi confirmed that it's self-driving car will use
the new Nvidia superchip future automotive self-
piloting capabilities.
Cognitive Computing , industry specific
applications
Collaborating and integrating IBM’s cognitive
computing technology with the HANA Cloud.
Self-driving car
Robotics, Cyber-Secuirty and IoT Self-driving car
26