Sei sulla pagina 1di 65

Introducing the SP Theory

An Intelligent machine
Results

Smart Machines and the SP Theory of Intelligence

Maverick Chardet

cole Normale Suprieure de Lyon


Colloquium ENS Lyon, 2015

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Simplicity and Power

The objectives of every scientic theory.


Power rst...
And for equivalent power, simplicity.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Simplicity and Power

The objectives of every scientic theory.


Power rst...
And for equivalent power, simplicity.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Simplicity and Power

The objectives of every scientic theory.


Power rst...
And for equivalent power, simplicity.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

A New Theory of Computing


Turing-equivalent, but more addapted to today's challenges:
Articial intelligence.
Big data.
Energy eciency.

ICMAUS: information-compression-by-multiple-alignmentunication-and-search
Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

A New Theory of Computing


Turing-equivalent, but more addapted to today's challenges:
Articial intelligence.
Big data.
Energy eciency.

ICMAUS: information-compression-by-multiple-alignmentunication-and-search
Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

A New Theory of Computing


Turing-equivalent, but more addapted to today's challenges:
Articial intelligence.
Big data.
Energy eciency.

ICMAUS: information-compression-by-multiple-alignmentunication-and-search
Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

A New Theory of Computing


Turing-equivalent, but more addapted to today's challenges:
Articial intelligence.
Big data.
Energy eciency.

ICMAUS: information-compression-by-multiple-alignmentunication-and-search
Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

The Objective

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Lossless Compression

Denition
Lossless data compression is a class of data compression algorithms
that allows the original data to be perfectly reconstructed from the
compressed data.
(source: Wikipedia, 05/2015)

Example
By using short codes for long patterns: cole Normale Suprieure
de Lyon -> ENSL.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Lossless Compression

Denition
Lossless data compression is a class of data compression algorithms
that allows the original data to be perfectly reconstructed from the
compressed data.
(source: Wikipedia, 05/2015)

Example
By using short codes for long patterns: cole Normale Suprieure
de Lyon -> ENSL.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Compression Is the Key

Facilitating the storage and management of data.


Economies in the transmission of big data.
Can be used as a way of computing and reasoning, in a more
ecient way in the use of energy.
Used to manage errors and uncertainty in data.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Compression Is the Key

Facilitating the storage and management of data.


Economies in the transmission of big data.
Can be used as a way of computing and reasoning, in a more
ecient way in the use of energy.
Used to manage errors and uncertainty in data.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Compression Is the Key

Facilitating the storage and management of data.


Economies in the transmission of big data.
Can be used as a way of computing and reasoning, in a more
ecient way in the use of energy.
Used to manage errors and uncertainty in data.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Compression Is the Key

Facilitating the storage and management of data.


Economies in the transmission of big data.
Can be used as a way of computing and reasoning, in a more
ecient way in the use of energy.
Used to manage errors and uncertainty in data.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Inspired by the Human Brain

The memory and the neural structure of the brain is modelled


by old and compressed data.
The information that comes to the brain via our senses is
modelled by new and uncompressed data.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Background and Objectives


Computing by Compression

Inspired by the Human Brain

The memory and the neural structure of the brain is modelled


by old and compressed data.
The information that comes to the brain via our senses is
modelled by new and uncompressed data.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Smart Machines?
Quote
While traditional computers must be programmed by humans to
perform specic tasks, cognitive systems will learn from their
interactions with data and humans and be able to, in a sense,
program themselves to perform new tasks.
J. E. Kelly and S. Hamm, Smart machines: IBM's Watson and the
era of cognitive computing, p7.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

From Pattern Matching...


Discovering generative grammars from data.
Parsing data using a grammar.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

From Pattern Matching...


Discovering generative grammars from data.
Parsing data using a grammar.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

...To Multiple Alignment

Similar to the Multiple Alignment concept in Bioinformatics


(with some modications).
Good alignement: when there are a lot of hits (positive
matches between symbols) and few and short gaps.
More exible, allows to parse incomplete information for
instance.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

...To Multiple Alignment

Similar to the Multiple Alignment concept in Bioinformatics


(with some modications).
Good alignement: when there are a lot of hits (positive
matches between symbols) and few and short gaps.
More exible, allows to parse incomplete information for
instance.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

...To Multiple Alignment

Similar to the Multiple Alignment concept in Bioinformatics


(with some modications).
Good alignement: when there are a lot of hits (positive
matches between symbols) and few and short gaps.
More exible, allows to parse incomplete information for
instance.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

SP Version of Multiple Alignment

Two categories of patterns to be aligned: new information and


old information.
Good alignement: leads to a large amount of compression of
new patterns in terms of old patterns.
An old pattern may appear several times.
No mismatches between old patterns.
In the future: generalisation in 2D.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

SP Version of Multiple Alignment

Two categories of patterns to be aligned: new information and


old information.
Good alignement: leads to a large amount of compression of
new patterns in terms of old patterns.
An old pattern may appear several times.
No mismatches between old patterns.
In the future: generalisation in 2D.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

SP Version of Multiple Alignment

Two categories of patterns to be aligned: new information and


old information.
Good alignement: leads to a large amount of compression of
new patterns in terms of old patterns.
An old pattern may appear several times.
No mismatches between old patterns.
In the future: generalisation in 2D.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

SP Version of Multiple Alignment

Two categories of patterns to be aligned: new information and


old information.
Good alignement: leads to a large amount of compression of
new patterns in terms of old patterns.
An old pattern may appear several times.
No mismatches between old patterns.
In the future: generalisation in 2D.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

SP Version of Multiple Alignment

Two categories of patterns to be aligned: new information and


old information.
Good alignement: leads to a large amount of compression of
new patterns in terms of old patterns.
An old pattern may appear several times.
No mismatches between old patterns.
In the future: generalisation in 2D.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Example of Multiple Alignment

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Astounding Figures
Human brain:

20 billion neurons.
20 watts of energy.
The size of a grapefruit.
A conventional computer with comparable cognitive capacity:

A gigawatt of electricity.
The size of a football eld.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Astounding Figures
Human brain:

20 billion neurons.
20 watts of energy.
The size of a grapefruit.
A conventional computer with comparable cognitive capacity:

A gigawatt of electricity.
The size of a football eld.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Optimising Processing

In the SP machine: processing = compression of information


using matching and unication.
Optimising compression leads to better processing and faster
search.
The SP computer uses its dierent levels of compression,
along with heuristics, as a guide to search.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Optimising Processing

In the SP machine: processing = compression of information


using matching and unication.
Optimising compression leads to better processing and faster
search.
The SP computer uses its dierent levels of compression,
along with heuristics, as a guide to search.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Optimising Processing

In the SP machine: processing = compression of information


using matching and unication.
Optimising compression leads to better processing and faster
search.
The SP computer uses its dierent levels of compression,
along with heuristics, as a guide to search.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Data-Centric Computing

To take inspiration from the human brain, the processing and


the memory should be more closely integrated.
Storing data is the same as computing for an SP machine.
To implement this system, two possible solutions are being
considered:

Computing with light.


Computing with DNA.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Data-Centric Computing

To take inspiration from the human brain, the processing and


the memory should be more closely integrated.
Storing data is the same as computing for an SP machine.
To implement this system, two possible solutions are being
considered:

Computing with light.


Computing with DNA.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Data-Centric Computing

To take inspiration from the human brain, the processing and


the memory should be more closely integrated.
Storing data is the same as computing for an SP machine.
To implement this system, two possible solutions are being
considered:

Computing with light.


Computing with DNA.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Data-Centric Computing

To take inspiration from the human brain, the processing and


the memory should be more closely integrated.
Storing data is the same as computing for an SP machine.
To implement this system, two possible solutions are being
considered:

Computing with light.


Computing with DNA.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

Unsupervised Learning
Computing in a Dierent Way

Data-Centric Computing

To take inspiration from the human brain, the processing and


the memory should be more closely integrated.
Storing data is the same as computing for an SP machine.
To implement this system, two possible solutions are being
considered:

Computing with light.


Computing with DNA.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Done

A lot of problems have been overcome during the rst


implementations.
The 1D multiple alignment framework is working.
A working and stable version has been published (SP70).

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Done

A lot of problems have been overcome during the rst


implementations.
The 1D multiple alignment framework is working.
A working and stable version has been published (SP70).

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Done

A lot of problems have been overcome during the rst


implementations.
The 1D multiple alignment framework is working.
A working and stable version has been published (SP70).

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

To Be Done

Improving the global framework.


Mathematics.
Graphical User Interface.
Developping an SP machine.
...

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

To Be Done

Improving the global framework.


Mathematics.
Graphical User Interface.
Developping an SP machine.
...

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

To Be Done

Improving the global framework.


Mathematics.
Graphical User Interface.
Developping an SP machine.
...

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

To Be Done

Improving the global framework.


Mathematics.
Graphical User Interface.
Developping an SP machine.
...

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

To Be Done

Improving the global framework.


Mathematics.
Graphical User Interface.
Developping an SP machine.
...

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Outline

Introducing the SP Theory


Background and Objectives
Computing by Compression

An Intelligent machine
Unsupervised Learning
Computing in a Dierent Way

Results
What Is Done
What Is Yet To Be Done
Potential applications

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Potential Applications

Automatic learning of natural language grammars.


Medical diagnosis.
Processing streams of data.
Representation and processing of graphical or audio
information.
Achieving human-like exibility and adaptability in robots.
The semantic web.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Potential Applications

Automatic learning of natural language grammars.


Medical diagnosis.
Processing streams of data.
Representation and processing of graphical or audio
information.
Achieving human-like exibility and adaptability in robots.
The semantic web.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Potential Applications

Automatic learning of natural language grammars.


Medical diagnosis.
Processing streams of data.
Representation and processing of graphical or audio
information.
Achieving human-like exibility and adaptability in robots.
The semantic web.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Potential Applications

Automatic learning of natural language grammars.


Medical diagnosis.
Processing streams of data.
Representation and processing of graphical or audio
information.
Achieving human-like exibility and adaptability in robots.
The semantic web.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Potential Applications

Automatic learning of natural language grammars.


Medical diagnosis.
Processing streams of data.
Representation and processing of graphical or audio
information.
Achieving human-like exibility and adaptability in robots.
The semantic web.

Maverick Chardet

SP Theory of Intelligence

Introducing the SP Theory


An Intelligent machine
Results

What Is Done
What Is Yet To Be Done
Potential applications

Potential Applications

Automatic learning of natural language grammars.


Medical diagnosis.
Processing streams of data.
Representation and processing of graphical or audio
information.
Achieving human-like exibility and adaptability in robots.
The semantic web.

Maverick Chardet

SP Theory of Intelligence

Appendix

For Further Reading


J. Gerard Wol.
Unifying Computing and Cognition.

CognitionResearch.org.uk, 2006.
J. E. Kelly and S. Hamm.
Smart Machines: IBM's Watson and the Era of Cognitive
Computing.

Columbia University Press, 2013.


J. Gerard Wol.
Smart Machines and the SP Theory of Intelligence.
arXiv.

1401.1669v1 [cs.AI], 2014.

J. Gerard Wol.
The SP Theory of Intelligence and the SP Machine, In Brief.
2015.
Maverick Chardet

SP Theory of Intelligence

Appendix

To Be Done (Honest Version)

Generalisation of the multiple alignment framework in 2D.


Create a more user-friendly system for using variables and
types.
Improving the framework for unsupervised learning:

Example: No capability for learning recursive structures or


exploiting symmetries.
Take account of the eect of recency in learning.
Verifying the coherence of entities in time and space.
Mathematics.
Graphical User Interface.
Developping an SP machine.

Maverick Chardet

SP Theory of Intelligence

Appendix

Example of Multiple Alignment (Readable Version) - 1

Maverick Chardet

SP Theory of Intelligence

Appendix

Example of Multiple Alignment (Readable Version) - 2

Maverick Chardet

SP Theory of Intelligence

Potrebbero piacerti anche