Models of Associative Learning Flashcards
What was the dominant idea in the first half of the 20th Century?
That contiguity was the guiding principle of conditioning
What 3 ideas influenced the change in the 1970s
Kamin - Blocking
Rescorla - contingency
Wagner - Relative signal validity effect
What is Relative Signal Validity?
Wagner 1968 - The CS must be a valid picture of the US judged relative to other present cues.
What did Wagner’s first experiment show?
Group 3 got strongest conditioning to the tone because of the intermixed trials L-> nothing
What was the issue with Wagner’s first experiment?
Group 2 got twice as many US’s as the others - could be a case that the animal is habituated to the presence of the US. Also the 2nd and 3rd groups get more light exposure.
How did Wagner’s second experiment solve issues of the first?
The 2 groups: correlated and uncorrelated got the same number of US and no US trials: Group 1: T1+L-> US and T2+L-> no US
Group 2: half US half no US for both
What did Wagner’s second experiment show?
Strongest conditioning to the Light in Group 2 because it is at least as good a predictor as the tones are. It is present in all trial when the US is present. Group 1 don’t learn about the light because it is less valid of a predictor.
Who introduced attention to conditioning theory?
Mackintosh - changes in the amount of attention a CS captures affects learning.
What phenomenon demonstrates the idea of attention in conditioning
overshadowing - L and N compete for attention
what divided Kamin, Mackintosh and Rescorla-Wagner
Mackintosh: changes in how the animal processes the CS
Kamin, Rescorla-Wagner: Changes in how the animal processes the US
What was Kamin’s theoretical analysis
conditioning depends on the US being surprising to the animal
What does the Rescorla-Wagner model tell us
1972 - strength of the association between CS-US varies on a single dimension
(L −∑V): what is L
How strong an association can be supported by that US - Large L -> high survival value
what is ∑V
the expectation of the US, based on the total associative strength of all CSs present. animal combines all of them to generate a prediction
What was Mackintosh’s contribution to theory?
changes in importance of CS as a signal and changes in intensity of CS effect learning
which model formulises the notion of surprise?
the rescorla-wagner model
whats the purpose of aB in the model?
They regulate the rate of conditioning, otherwise all conditioning would happen in one trial
in wagners second experiment which group had more Light - US pairings?
They each had the same number of pairings (50% of the time)
what is the implication of Rescorla- Wagner model on real life?
natural conditioning probably always involves the conditioning of compounded cues - i.e. the context
How do we know information value is important?
from Blocking and contingency learning
What would make V reach the asymptote faster i.e. for the learning curve to be steeper
if the US was more surprising i.e. if (L-V) was larger.
What determines L (the asymptote)
the magnitude of the US
What would cause the steepness of the learning curve to decline as it reaches the asymptote?
the US becoming less surprising as CS predicts it, i.e. (L-V) is smaller
Calculate the change in associative strength of the light:
Phase 1: N -> US
Phase 2: L+N-> US
∑V= VL + Vn = 0 + 1
△VL=0.2 (1 −1)=0
zero because there is no surprise on the compound trial
what is unblocking?
Kamin changed the intensity of the US in the second phase leading to learning about the light. Positive change in associative strength of the light:
△VL=0.2 (2 −1)=+0.2
L is 2 now
What is the value of L during extinction?
zero.
What happens to the change in associative strength of the CS in extinction?
It decreases.
Phase 1: The CS is paired with the US until V=1
Phase 2: Extinction L=0
△Vn=0.2 (0 −1)=−0.2
Vn approaches an asymptote of zero
How can Pavlov’s conditioned inhibition be explained in terms of Rescorla-Wagner’s model?
Light-> US
L+N-> No US
Noise becomes an inhibitor its Vn goes below ZERO
VL decreases from 1
∑V approaches the zero asymptote
What does the compound in Pavlov’s conditioned inhibition paradigm predict?
Nothing because the light predicts the US just as much as the noise predicts no US, cancels out
what happens if an excitor and inhibitor are compounded during extinction?
the inhibitor may protect the excitor from total associative loss. Implication for exposure therapy: clinicians may extinguish excitors in the presence of cues that become inhibitory during exposure. The excitor alone would still elicit fear. Reason for relapse.
which would be more effective during extinction: an excitor paired alone to the US, an excitor-inhibitor compound paired to the US, or an excitor-excitor compound paired to the US?
excitor-excitor because: Greater decrease in associative strength when an excitor is extinguished in compound with another excitor.
○ △VL=0.2 (0 −(1+1))=−0.4
where ∑V = 1+1 because the excitors are paired with the US until their V=1 in the initial phase.
What is Kremer’s overexpectation effect?
Vn=1 and VL=1 for a shock US
N+L-> shock
they each lose associative strength due to overprediction:
△Vn=0.2 (1 −(1+1))=−0.2
once each has a V=0.5 then the reduction stops
what happens when a third stimulus is added in the second phase of Kremer’s overexpectation effect trial?
it becomes a conditioned inhibitor
- △Vx=0.2 (1 −(1+1+0))=−0.2 issue: the description of a conditioned inhibitor as signalling 'no US' is inaccurate here... better description = an inhibitor signals that an upcoming US is not as strong as other cues on the trial predict it to be.
what is the biggest message from the overexpectation effect?
CS-US pairings can actually reduce associative strength
How does the context lead to blocking in zero contingency?
A=Room
X=CS zero contingency
when we have AX-US, the US is not surprising because it is already predicted by the context, A
What happens when contextual conditioning is reduced in negative contingency training?
little inhibition is acquired by the CS
Which theories focus on contextual conditioning?
Rescorla-Wagner and Comparator theories
What did comparator theories add to contextual conditioning theory ?
The idea that CS-US can be stronger or weaker than context-US association. The comparison is made AFTER learning has occurred so it determines PERFORMANCE not learning about the CS
According to comparator theories what happens if you weaken the context-US association after learning about the CS occurs?
The subject responds more to the CS - shown in Ralph Miller’s lab.
But no one has shown the converse of strengthening the association to reduce responding to the CS.
How does the extinction of inhibition contradict the Rescorla-Wagner model?
Because if a conditioned inhibitor is presented in extinction trials with no US the model says it will REMOVE inhibition:
○ △Vn=0.2 (0 −(−1))=+0.2
but this is not possible since it can’t decrease inhibition if it is still being presented with NO US…
How does Latent inhibition challenge the Rescorla- Wagner model?
The model does not account for the fact that if a CS is preexposed before conditioning begins - conditioning responding appears relatively slowly
How do Rescorla-Wagner and Mackintosh differ in their explanations of blocking
R-W: emphasize the ineffectiveness of the US
Mackintosh: Emphasizes the ineffectiveness of the CS
How did Mackintosh’s blocking experiments contradict the R-W model?
Group 1: N-shock ————— LN-SHOCK
Group 2: N-shock LN-shock LN-SHOCK
R-W predicts that nothing should’ve happened during Phase 2 and groups should not differ because if N had already been conditioned to the asymptote then there should be no change in the value of V for either CS. BUT group 2 learned less about the light.
What occurred in Mackintosh’s blocking experiments?
Group 1: N-shock ————— LN-SHOCK
Group 2: N-shock LN-shock LN-SHOCK
Group 2: phase 2 interfered with learning because the animals recognised that L was a redundant predictor of shock and they learned to pay less attention to it so in Phase 3, they ignored it
Who introduced attention in conditioning
1970s Mackintosh - Attention depends on how well the CS predicts a US relative to other CS
How does the Mackintosh model explain latent inhibition?
during preexposure to the CS, the value of a goes down because it is no better than the context at prediction no US. So increases are then small when CS is paired with a US
How does the Mackintosh model explain the blocking experiment:
Group 1: N-shock ————— LN-SHOCK
Group 2: N-shock LN-shock LN-SHOCK
Group 2: Phase 2: first trial a is higher for noise as it has been established as a good predictor of the shock. a for the noise decreases as trials go on. By Phase 3 learning about the light is slower due to a low a.
What would contradict Mackintosh’s statement that the value of a can only change after the first conditioning trial?
if we saw complete blocking on the first trial (this would support the R-W Model)
How does the Hall-Pearce negative transfer contradict Mackintosh’s model?
- Group 1: Tone - shock Tone - SHOCK
- Group 2: L - shock Tone - SHOCK
Mackintosh says that for group 1, the tone receives a high a in phase 1 so should receive faster learning in phase 2. BUT Hall and Pearce found they had SLOWER learning.
- Group 2: L - shock Tone - SHOCK
what controls the speed of learning in the Rescorla-Wagner model
aB
How does L (the asymptote) differ for a compound ?
for a compound, the asymptote is lower (L, the maximum that can be learned) for each CS in the compound
For a compound to split uncertainty (L) equally, what needs to happen?
Their salience (a) must be equal so that they split L (uncertainty) equally
what happens if a is larger for one CS in a compound than the other?
The CS with a larger a has more salience and takes up more of the uncertainty (L), it reaches a higher associative strength