associative learning
DESCRIPTION
Associative Learning. Psychology 3906. Introduction. Every species tested seems to show some form of associative learning There are many possible responses Are they all served by some overarching ‘associative learning system?’. Constraints. - PowerPoint PPT PresentationTRANSCRIPT
Associative Learning
Psychology 3906
Introduction
• Every species tested seems to show some form of associative learning
• There are many possible responses• Are they all served by some
overarching ‘associative learning system?’
Constraints
• Early on (meaning the 1960s….) people looked at so called constraints
• Taste aversions• “Misbeahviour”• The whole notion of adaptive
specializations of learning• Not really a clear research program
More recently..
• The modular view has helped out here• Intelligence is modular• Memory, for example, has many
subsystems, we accept that easily• Same with perception• So learning may as well
What is learning?
• Some event at time 1 affects behaviour at time 2
• Probably the best definition out there, though there are others
• Bob Rescorla
What are we interested in?
• Conditions for learning• Contents of learning• Effects on behaviour
Conditions for learning
• Contingency• The “Rescorla control” vs conditioned
inhibition• Compounding• Features and blocking• Surprisingness
The Model:
• ΔVi – Si(Aj-Vsum)• i = CS• j = US• S = Salience• A = Value of the US• V = amount of conditioning• These quantities are, of course, hypothetical
An example
• OK, say a food pellet = 100• Say salience of a light CS = .2• Vsum = 0 (at the start of the experiment,
there is no conditioning yet
OK, now for the numbers
• Trial 1• ΔVi – Si(Aj-Vsum)
=.2(100 – 0)=20
• Trial 2ΔVi = .2(100-20)
=16
Continued….
• Trial 3ΔVi = Si(Aj-Vsum)ΔVi = .2(100-36)12.8
• And so on….• Less and less conditioning as time goes
by• Cool eh
More on conditions
• Gallistel’s model is more to do with duration of events and number of pairings
• Events have to “go together”– Preparedness– Belongingness– No model deals with that well at all
I just want to belong to something….
• Spatial and temporal contiguity can be considered types of belongingness
• Think about backwards conditioning, it never works
• When you look at these two types of contiguity from a functional viewpoint it makes a great deal fo sense
Prior learning affects new learning
Group Phase 1
Phase 2
Test Result
Control Nothing LT+ T CR
Blocking L+ LT+ T No CR
The contents of learning
• Is it an association between stimuli?• Is it an association between a stimulus
and a response?• It seems to depend on the preparation• How in the hell could you tell?• As usual, Bob Rescorla figured this
out…
S-S and S-R
US
CS
UR
If we could just get rid of that US – UR bond…..
US
CS
UR
Rescrola (1973)
• So, how do you get rid of a response that is hard wired to a stimulus?
• Well, if you use CER, then your response is startle right?
• How do you get rid of a startle reflex?• Habituation!!• (Bob is a smart man)
Design
Group Phase 1 Phase 2 Test
Habituation
L -> N Noise Light
Control L -> N Nothing Light
Results
• Less suppression in Habituation group• (In other words, more responding)• Therefore, the connection MUST be S –
S• WOW!
But, after all of that….
• Need it be a connection?• Gallistel says that it is contingency itself
that is learned• Different features of the CS and US are
learned• Holland’s experiment
Effects on Behaviour
• Learning vs. performance• The Behaviour system approach• Are Pavlovian and Instrumental learning
different?• Operationally of course yes• But, what about occasion setting and
within event learning
In sum
• It may simply be that most types of what we call associative learning use many of the same modules
• Cause effect realationships• Patterns• Predicting the future is pretty adaptive