Machine Learning

jamaicacooperativeAI and Robotics

Oct 17, 2013 (3 years and 7 months ago)

80 views

Machine Learning

Example sky AirTemp Humidity Wind Water Forecast EnjoySport

1 Sunny Warm Normal Strong Warm Same Yes

1 Sunny Warm High Strong Warm Same Yes

1 Rainy Cold High Strong Warm Change No

1 Sunny Warm High Strong Cool Change Yes

What concept do you learn from these
examples?

Credit History of Loan Application

No. Risk Credit History Debt Collateral Income

1. high bad high none 0k


15k

1. high unknown high none 15k


35k

1. high unknown low none 15k


35k

1. high unknown low none over 35k

1. high bad high none 0k


15k

1. moderate good low none over 35k

1. low good high none over 70k

1. high bad high none 15k
-
35k

What risk does a person with unknown credit history, low credit, high
income have?

Arch examples

Arch

Arch

Near miss

Near
miss

How does machine
recognize an arch? What
concept should it learn in
order to recognize an arch?

Learning

Herbert Simon defines learning as:


any change in a system that allows it to perform better the second
time on repetition of the same task or on another task drawn from
the same population (1983)



Involves generalization from experience.



Examine only a fraction of all possible examples.



Must acquire knowledge to use it in the future example

The problem of induction

Induction Biases

The available data are not sufficient to guarantee optimal
generalization, no matter what algorithm the learner uses.

Learning algorithm must generalize heuristically; they must
select those aspects of their experience that are most likely to
prove effective in the future.


Inductive Biases

Concept of Arch

Arch

Part

Part

Part

brick

brick

brick

supports

supports

Representation Language: Input data: graph

Pyramid arch

Arch

Part

Part

Part

brick

brick

pyramid

supports

supports

Background Knowledge

Polygon

brick

pyramid

isa

isa

Generalization

Arch

Part

Part

Part

brick

brick

polygon

supports

supports

Operations: generalizing a network by replacing
node or links names with a more general concept

Description of A near miss

Arch

Part

Part

Part

brick

brick

polygon

supports

supports

touch

touch

Specialization: Excluding the near miss

Arch

Part

Part

Part

brick

brick

polygon

supports

supports

must
-
not
-
touch

must
-
not touch

Operations: specializing a network by adding links