1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural...

10
CVG-UPM COMPUTER VISION Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual Campoy Computer Vision Group C.A.R. - U.P.M. 2 CVG-UPM COMPUTER VISION Machine Learning and Neural Networks P. Campoy P. Campoy P. Campoy Intelligence and learning What is intelligence? What are intelligent machines? The learning relevance Building intelligent machines Objectives of the subject Applications

Transcript of 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural...

Page 1: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Machine Learning & Neural Networks

1.- Intelligence & Learning

byPascual Campoy

Computer Vision GroupC.A.R. - U.P.M.

2

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Intelligence and learning

What is intelligence?

What are intelligent machines?

The learning relevance

Building intelligent machines

Objectives of the subject

Applications

Page 2: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

3

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

¿ What is intelligence ?

“Intelligence is a very general mental capabilitythat, among other things, involves the ability toreason, plan, solve problems, think abstractly,comprehend complex ideas, learn quickly andlearn from experience. […], it reflects a broaderand deeper capability for comprehending oursurroundings -- "catching on," "making sense" ofthings, or "figuring out" what to do”statement signed by 52 experts in intelligence in the WallStreet Journal & in the Intelligence Journal 1994

4

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

¿ What is intelligence ?

•“Prediction is the real essence of cerebralfunction”, Rodolfo Llinás “I and the vortex”

Recognition, prediction and learning

an intelligent machine has to recognize thesituation and predict the future based on

learned previous experiences

•“Intelligence is the capability of predictingfuture”, Jeff Hawkins “On Intelligence”

Page 3: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

5

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

¿ Machine intelligence ?

6

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

¿ Machine intelligence ?

Alan Turing test:• original version: “a machine is said to beintelligent if it can fool a judge as often as aperson in pretending to be a man/woman”• later extended version: “a machine is saidto be intelligent if it can fool a judge to be aperson”

Page 4: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

7

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

8

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

¿ Machine intelligence ?

•“Chinese room paradigm: suppose that amachine convinces a human Chinese speaker thatthe program is itself a human Chinese speaker, thenthe machine is substitute by a non Chinese speakingperson in a room provided with the same algorithmand information as the machine had. Does he/sheunderstand Chinese? ” de John Searle

Page 5: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

9

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

10

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Learning relevance

imagen sólo para la imagen sólo para la mitad izquierda del mitad izquierda del

auditorioauditorio

imagen sólo para la imagen sólo para la mitad derecha del mitad derecha del

auditorioauditorio

Page 6: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

11

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

learning relevance

12

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

¿How to build intelligent machines?

Explicit model− top-down methodology

Example learning− boton-up methodology

- paradigms: physical equations expert systems

- paradigms: experimental parameters neural networks

Self-learning

Parallel processing

Distributed Patterns

Signal processing

Programmed

Serial Processing

Symbolic data

High dimensional statespace searching

Learning based systemsKnwoledge based systems

Page 7: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

13

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Ojectives of the subject

Find a model that predicts the right output to anew input, considering outputs to previous inputs.

y

x?

?

x1

x2

14

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Learning applications

Pattern Recognition

Control

Identification

Associative memory

Optimization

Page 8: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

15

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Pattern Reconition Applications

Representation, recognition and predictionof the state of a multidimensional inputvector, which can be:- The state of an industrial process from the field

sensors (includes also alarm management)- The state of an electrical net- The atmospheric state from a spread net of sensors- The seismical or geological state- The state of a vehicle (e.g. airplane, car, …)- The commercial situation of a company from its whole

accountability- Patterns in images (e.g. visual, IR, electromagnetic, ..)

16

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Computer Vision applications

Abnormalcristalitation

preprocessing segmentation feature extraction Recognition

Page 9: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

17

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Your applications

1. Synthetic data

?

2. Real data

18

CVG-UPM

CO

MP

UT

ER

VIS

ION

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Homeworks in Moodle:Files/Exercises

Page 10: 1.- Intelligence & Learning - UPM · CVG-UPM COMPUTER VISION P. Campoy Machine Learning and Neural Networks Machine Learning & Neural Networks 1.- Intelligence & Learning by Pascual

19

CVG-UPMC

OM

PU

TE

R V

ISIO

N

Machine Learning and Neural NetworksP. CampoyP. CampoyP. Campoy

Exercise 1.1

Print the 2 different handwritten images of thesame number for 4 different numbers

4

7

8

9