Artificial Intelligence - A Brief Scan

Artificial Intelligence (AI) is the science and engineering of making intelligent machines.
Machine learning (ML) is a subfield of AI that explores the study and construction of
algorithms to learn from and make predictions on data without being explicitly
programmed. Deep learning (DL), one of the most promising approaches in AI, uses
artificial neural networks (ANNs) that consist of one of more hidden layers.
DEFINITIONS

3
SG-Tech Slides
Machine
Learning
Use algorithms such as
Single Vector Machines
CA Solutioning
Statistical approaches that
do not improve artificial
intelligence over time.
Artificial
Intelligence
Symbolic logic machines (rules
engines, expert systems) are
non ML AI.
Computer Assisted
Solutioning
AI
ML
DL
Deep Learning
ANNs with one or more
hidden layers powered by
both non DL and DL-specific
algorithms
COMMON TERMS
Specific Deep
Learning
DL architecture and algorithms set
up for specific tasks, e.g. image
recognition
Specific DL

4
SG-Tech Slides
Exponential increases in computing power has meant that
certain algorithms and architecture which were
impractical in the past, e.g. DL, have become viable.
Computing Power
Access to large amounts of data available in the cloud, has
made DNNs feasible.
Big Data
Recent AI innovations have been powered by combining
algorithms and architecture together in novel ways to
solve specific sets of problems.
Algorithms and Architecture
3 Factors
Determines AI’s Potential

5
SG-Tech Slides
Artificial Narrow Intelligence
Designed to solve highly specialized
problems.
Computing Power
With the slowing down of Moore’s
law, quantum computing offers the
possibility of brute force advances
based on current approaches
New Algorithms/Architectures
Focused on speed, accuracy as well
as reducing the amount of data
needed
Processing Speed
Dedicated AI chips, Parallel
processing and in-situ computing
solutions can offer increases in
processing speed for AI
POSSIBLE TRAJECTORIES
In AI Research

6
SG-Tech Slides
Genetic Algorithms
Using AIs to build AIs more
effectively
2040 ?
AI as a Discipline
Current approaches to AI are
largely empirical – i.e. we don’t
really understand why algorithms
work the way they do.
BNN Emulation
New areas of research include
more closely studying and
emulating biological neural
networks, including physical or
virtual embodiment.
Artificial General Intelligence
An AGI will be able to generate and
merge multimodal datasets, form
“situations” based on concepts, and
make and test hypotheses across
different and new areas.
POSSIBLE TRAJECTORIES
In AI Research

7
SG-Tech Slides
Rules-based
Improve productivity and customer
experience
Firm specific
Change business model
Ubiqitous
Small AI
E.g. Amazon 1-click
ML/DL
Improve organizational intelligence
Problem specific
New business model(e.g.MLAAS)
5-10%
Big AI
E.g. Watson
TYPES of AI
Applicable to Industry

SG-Tech Slides
% Over the next 5 years
CAGR
In Millions over 15 Major
Economies
Job
Losses In $billions in 2015
VC
Funds
Market Size
The AI industry in $billion by 2020
13.7
IMPACT of AI
33.7 5 1.2

9
SG-Tech Slides
Algorithmic trading, fraud detection, credit
risk assessment, identity management
Web search, solutioning, prediction, design
Image recognition, conversational agents,
personalisation
Finance
Research
Business Models
High speed threat identification and
intervention
Medical imaging, patient monitoring, risk
assessment, diagnostics and drug discovery
Law - Legal research, e-discovery, outcome
prediction and contract analytics.
Journalism – News writing, news discovery.
Education – Artificial teaching assistants
Creative Services – Game design, art, music
Cybersecurity
Healthcare
Others
IMPACT of AI
On Research, Business Models and Industries

10
SG-Tech Slides
24k+
Publications
11
Of top 12 AI Coys
USA
$4.2 billion in VC funding (2015). 499 AI
companies
40k+
Publications
1
Of top 12 AI Coys
CHINA
Leader in visual and speech recognition.
2nd
In FWCI rankings
6
Local AI startups
SINGAPORE
Top researcher (by FCWI) is ranked 33rd in the
world. Institiutions are ranked between 170s and
1500s.
SINGAPORE’S
Competitive Advantage

12
SG-Tech Slides
Single layer ANN
Pattern
Identification
1
2 layer DNN
Building
Layers
3
Supervised learning with
labelled data
Training
4
Testing with new data.
Testing
5
Using sparse coding for feature
extraction
Feature
Extraction
2
DEEP LEARNING
A Simulation

Artificial Intelligence - A Brief Scan

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Artificial Intelligence - A Brief Scan

Similar to Artificial Intelligence - A Brief Scan (20)

Recently uploaded

Recently uploaded (20)

Artificial Intelligence - A Brief Scan