WalsheACT-R

Abstraction and ACT-R
Outline
Motivations for the Theory
– Architecture
– Abstraction
Production Rules
ACT-R Architecture of Cognition
Architecture
• There is a distinction between brain
architecture and cognitive architecture.
• This is a difference between the
physical description of the system, and
the functional properties that it enables.
• Some examples…
• The common feature to each is in the
method of abstraction.
– Moving from a more complex, lower level
system of interaction, to a higher level
language in which those lower level
features are typed.
Some Characteristic features of Abstraction.
Constraints on the Abstraction
• Typing
– Lower level features are given higher level types
• Simplification
– The higher level abstraction is obligated to
simplify conceptual complexity
• Morphism
– There must be a mapping.
– ‘Many to one’
Production Systems
Comprise a language with sentences of the form
If <<condition>> then <<event/action>>
Developed by Emil Post (1943), also known as ‘Post
Production Rules’
Turing Complete
Solve a Linear Equation
• If the expression has X = N, where N is a number, THEN
halt and check by substituting N in the original equation
• If there is a term in X on the right-hand side, THEN subtract
it from both sides, and collect the terms.
• IF there is a numerical term on the left hand side, THEN
subtract it from both sides, and collect the terms
• IF the equation has the form NX = M, N = 1, THEN divide
both sides by N
•
•
•
•
If the expression has X = N,
where N is a number, THEN
halt and check by substituting
N in the original equation
If there is a term in X on the
right-hand side, THEN
subtract it from both sides,
and collect the terms.
IF there is a numerical term
on the left hand side, THEN
subtract it from both sides,
and collect the terms
IF the equation has the form
NX = M, N = 1, THEN
divide both sides by N
•
•
•
•
X = 2X + 5
X – 2X = 2X – 2X + 5
-X = 5
X = -5
ACT-R
(Adaptive Control of Thought-Rational)
• Modules
• Buffers
• Productions
Modules - Fodor (Modularity of Mind)
• Features
– Functional
separation
Domain specific
• Interaction between
modules
Buffers
• Active Content
• Interface
• Interaction between
modules
– Provides the link
Productions
• If <<condition>>
then <<result>>
• State Change
– Contents of Buffer
– Contents of
Modules
Learning
• Adaptive control
– Update
• Goal structures
• Declarative memory contents
• Productions
Structure of Declarative Module
‘Chunks’ – A collection of related
information.
Action023:
isa chase
agent dog
object cat
Fact3+4:
isa addition-fact
addend1 three
addend2 four
sum seven
Summary
• Cognitive Science and Act-R benefit by
engaging in abstraction
• By matches conditions set by contents
in the buffer, and specifying to changes
in that content. Act-R models
cognition.