Scribd
Upload
6 views

Week 2 Characterization of Learning Problems: Nptel Video Course On Machine Learning

Uploaded by

Sujithra Jones
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
6 views

Week 2 Characterization of Learning Problems: Nptel Video Course On Machine Learning

Uploaded by

Sujithra Jones
Copyright
© © All Rights Reserved
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 17

NPTEL

Video Course on Machine Learning

Professor Carl Gustaf Jansson, KTH

Week 2 Characterization of Learning Problems

Video 2.2 Objects, Categories and Features


The lecture will discuss the following items:

Objects and Features


Object space and Object Language

Category and Category structure


Category space and Hypothesis Language
Object and Feature
Synonyms Synonyms

Thing Property
Entity Attribute
Observation Characteristic
Data, Data-item

Record , Tuple Field


Row, Vector Column
Variable, Output Variable
Instance, training instance Independent variable, Predictor Variable
Example,training example Target or Category feature

The Object (Data, Observation) Language is the chosen language (formalism) in which
objects and features can be described.
Types of Features

Ordinal (binary)
Discrete numerical (integers)
Continous numerical (real numbers)
Symbolic
Structural (e.g graphs or lists)
Example from the ZOO Dataset: buffalo,1,0,0,1,0,0,0,1,1,1,0,0,4,1,0,1,1
animal_name buffalo category symbolic feature

Hair 1 predictor ordinal feature


Example of Feathers
Eggs
0
0
predictor ordinal feature
predictor ordinal feature
Object or Milk 1 predictor ordinal feature
Airborne 0 predictor ordinal feature
Feature vector Aquatic 0 predictor ordinal feature
Predator 0 predictor ordinal feature
Toothed 1 predictor ordinal feature
The Object Backbone
Breathes
1
1
predictor ordinal feature
predictor ordinal feature
Language in this Venomous 0 predictor ordinal feature
Fins 0 predictor ordinal feature
case is the Legs 4 predictor ordinal feature
specific formalism Tail 1 predictor discrete numerical feature
Domestic 0 predictor ordinal feature
for specifying Catsize 0 predictor ordinal feature
Feature vectors
class_type 1 category discrete numerical feature
Object space

Instance space
Population

Subsets of object space available for Learning

Sample, Training sample, Statistical sample


Data-set, Table, Array
Training example set
Example from the ZOO Dataset

The Object space or population is the set of all potential


feature vectors with feature values as can be
expressed in the ZOO object language.

The Sample or Data-set is the whole set of ZOO feature


vectors.

The Extension of the Concept of buffalo is the set of all


buffalos in real life.
Category and Hypothesis (Concept, Category)
Space
Synonyms

Concept Intension of Concept


Class Schema
Hypothesis Model
Target function Classifier
Type Kind of thing

The Hypothesis (Category, Concept) Language can theoretically


be different from the Object language, but is typically syntax-wise consistent with
the latter. The syntax for feature values have to be extended with suitable
generalizations of the normal values.
Normally available background knowledge are embedded in the Hypothesis
Language with the purpose to constrain the learning process (Language Bias).

The Hypothesis space is the space spanned by what can be expressed in the
hypothesis language.
Generalization
Object Category
Definition
Instance-of
Element-of

Subset of the Data-set Subset of the


Consistent with Subset-of Object space
the category consistent with
definition the category
definition
Example from the ZOO Dataset

Example of fish 0,0,1,0,0,1,?,1,1,0,?,1,0,1,?,?,4


Conception Generalization
Definition tuna, 0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,1,4
stingray, 0,0,1,0,0,1,1,1,1,0,1,1,0,1,0,1,4
seahorse, 0,0,1,0,0,1,0,1,1,0,0,1,0,1,0,0,4
pike, 0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,1,4
The Hypotheses piranha, 0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4
Language is the herring,
haddock,
0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4
0,0,1,0,0,1,0,1,1,0,0,1,0,1,0,0,4
same as the Object dogfish, 0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,1,4
chub, 0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4
Language apart from the catfish, 0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4
carp, 0,0,1,0,0,1,0,1,1,0,0,1,0,1,1,0,4
introduction of a wildcard (?) bass, 0,0,1,0,0,1,1,1,1,0,0,1,0,1,0,0,4
for ordinal feature values.
Category Structures

Persistent and domain relevant category structures

Taxonomy
Class structure, class hierarchy, class lattice
Concept structure, Concept hierarchy, Concept lattice
Type structure, Type hierachy, Type lattice

Temporary categorystructures
used to support the learning process

e.g. Version spaces.


Category 1

Specialization Generalization

Category 2

Specialization Generalization

Category 3
Concept Hierarchy
Category 1
Generalization
Generalization
Category 2 Category 3 Category 4

Generalization Generalization

Category 5 Category 6 Category 7 Category 8 Category 9 Category 10


Concept Lattice
Category 1
Generalization
Generalization
Category 2 Category 3 Category 4

Generalization Generalization

Category 5 Category 6 Category 7 Category 8 Category 9 Category 10


The ZOO example extended
The complete Taxonomy for the Buffalo

Animalia

Bilateria animals with bilateral symmetry


Deuterostomia animals that share certain features of embryonic development
Chordata animals having a notochord at some stage of their development
Vertebrata animals with backbones
Gnathostomata vertebrates that have jaws
Tetrapoda gnathostomata with four legs.

Mammalia

Theria mammals that give birth to live kids.


Eutheria non marsupials
Laurasiatheria hoofed animals which bear most weight on one toe
Ruminantia a chewing mammal – an ungulate
Bovidiae cattle
Bovinae a middle to large size ungulate
Bovina Bison or Bos
Bison
Bison Bison = Buffalo
The attribution of features to different levels of a category structure
When a conceptual structure is formed during a learning process, features will attributed to the
categories of different generality.

More specialized categories INHERIT the features of more general categories.

Features are normally not spread evenly across the abstraction levels, most features are grouped
in the mid range of the conceptual structure, often called the BASIC LEVEL.

Example:

Fruit Apple Delicius


Mackintosh
Peach Freestone peach
Cling Peach
Grape Concord
Green seedless

BASIC
LEVEL
NPTEL

Video Course on Machine Learning

Professor Carl Gustaf Jansson, KTH

Thanks for your attention!

The next lecture 2.3 will be on the


topic:

Feature Related Issues

You might also like