19
eCognition Model Components

eCognition Model Components

  • Upload
    osborn

  • View
    56

  • Download
    4

Embed Size (px)

DESCRIPTION

eCognition Model Components. Model Components. Variables and Constants Operators and Functions Links Logical States Messages Change Storage. A Simple Example. a + b = c. If this is part of a larger model, how is it controlled? What does the EQUALS mean? - PowerPoint PPT Presentation

Citation preview

Page 1: eCognition Model Components

eCognition Model Components

Page 2: eCognition Model Components

2. eCognition Model Components 2

Model Components

• Variables and Constants

• Operators and Functions

• Links

• Logical States

• Messages

• Change Storage

Page 3: eCognition Model Components

2. eCognition Model Components 3

If this is part of a larger model, how is it controlled?What does the EQUALS mean?How many ways should we be able to use this piece of knowledge?

a + b = c

A Simple Example

Page 4: eCognition Model Components

2. eCognition Model Components 4

Instead of writing programs, turn the knowledge itself into a computing machine

Programmatic:if input(a) and input(b) then c = a+belse if input(a) and input(c) then b = c-aelse…

a + b = c

A new approach to knowledge

Structure

NYK - Not Yet Known

Page 5: eCognition Model Components

2. eCognition Model Components 5

Values flow through the structure in any direction.

The EQUALS operator allows for logical control.

a + b = c

Use the structure itself

Page 6: eCognition Model Components

2. eCognition Model Components 6

Information Coming the Other Way

Example: a + b = c

We have a range on one variable, producing a range on another.

Page 7: eCognition Model Components

2. eCognition Model Components 7

Is It Just Numbers

The structure can propagate a wide range of entities through its connections, and the operators can be operating on analytic or experiential information, and the structure can be changing itself - it is a lot more than just numbers.

Page 8: eCognition Model Components

2. eCognition Model Components 8

Variables and Constants

Every variable and constant has a logical state and may have a value - logical, numerical, string, list, object.

Every variable and constant is a linkable object - it can support unlimited connections.

Variables are addressable by a name which can have unlimited depth of context and multiple context -

Model.Animal.Mammal.Whale.MobyDick

Page 9: eCognition Model Components

2. eCognition Model Components 9

Operators & Functions

The distinction between the two in the network is blurred

• many functions are invertible - ABS(X) • many functions are actually pieces of network machinery - WHILE, GETPUT, SEQUENCE

Operators have a fixed orientation to their connections (the number of connections may be variable - the PLUS operator can function with up to 1024 connections).

Page 10: eCognition Model Components

2. eCognition Model Components 10

Operators & Functions

The network has operators that are not obvious in the text

• SPINE and LEVEL operators allow logical structuring of network text

• INDEX operator to represent an index into a list• structure to represent potential inferences

Some textual representations convert to simpler forms

• Minus is a Plus operator with different orientation• a + b + c is held in the network as one operator

Page 11: eCognition Model Components

2. eCognition Model Components 11

Links

Links connect operators and variables together.

They store the information produced by operators, allowing an operator to have many concurrent outputs.

Links provide for bi-directional flow of information.

Operators can add links to themselves for storage of states, or an operator’s only purpose can be to add links to other operators - LISTLINK.

Page 12: eCognition Model Components

2. eCognition Model Components 12

Logical States

Logical states range over

• Not Yet Known• Unknowable - Bayesian values or existence• Error• False• True

and control the phasing of operations

R unningN YK U KE D oesn’tExist

BayesianValues

Valid

Error

True

False

Page 13: eCognition Model Components

2. eCognition Model Components 13

Messages

Messages can be • simple singular values - True, 2.3, “Fred”• objects - Joe, MobyDick, Glock• lists - {1,2,3}• Bayesian values - UKE 0.7• signalling nonexistence of structure • alternative values using list transmission - 2..4, 3<->6• structure - a < b, IF X > Y THEN P < Q

There is no conceptual limit on the size of a message, as it is constructed out of the same elements that make up the network

Page 14: eCognition Model Components

2. eCognition Model Components 14

Self Modification

X = + ListLink =

B

CA%

D

A%ListLink

X = + ListLink A%

X = +

B

C

D

B=9

C=3X=5

D=-7

+

Page 15: eCognition Model Components

2. eCognition Model Components 15

Change Storage

All changes to the network can be stored in memory in a layered fashion for later retrieval.

This allows values to be changed or hypothetical structure to be generated, its operation observed, and then the scenario to be undone.

A change can be attempted, and if successful, its changes merged with a previous store level which is still tentative.

Page 16: eCognition Model Components

2. eCognition Model Components 16

Functions and Operators

• eCognition provides a comprehensive set of predefined operators and functions.

• The operators in the network represent the elemental or atomic level of analysis

• Each operator determines its own activity, so the network is micro-scheduling what it does.

• The user can add script operators, and user-defined functions.

Page 17: eCognition Model Components

2. eCognition Model Components 17

Handling Uncertainty

• eCognition is designed to handle uncertainty – it can reason using partial or fuzzy knowledge.

• Variables can hold tentative knowledge – alternative values (ranges for numbers) or distributions.

• Logical variables can support values between False and True (0 to 1 is used, with 0.5 representing Unknowable). These intermediate values can be overridden, but True and False cannot be overridden.

Page 18: eCognition Model Components

2. eCognition Model Components 18

How It Works

• Setting the value of a variable causes the new value to propagate in all directions through the network, until there are no new paths to propagate

• Ranges of numbers, alternative strings and lists are also propagated

• Some operators alter their local topology and then destroy themselves

• Spreading activation is also used - one operator forces another to become active, as the message is too complex to propagate

Page 19: eCognition Model Components

2. eCognition Model Components 19

Logical Control

• Everything has a logical state

• Every statement or equation lives in a logical “block”, or environment, and is controlled by a “head” variable

• A head variable can turn its block on, make it false or unknown, or make it go dark

• A head variable can be tested for truth (higher structures are invertible too)

• Logical control allows the user, and the model itself, to control which part of the structure is active