Rescorla & Wagner model & Wagner SOP model Flashcards
what gets smaller with each trial
change is associative strength
Blocking effect
happens on first trial of stage 2
e.g stage 1 is tone stage 2 is light
tone which reached asymptote by end of stage 1 has blocked learning about the light
Mackintosh argued…
that RWM was wrong because he demonstrated that blocking did not occur with only one stage-2 trial. Instead he argued that that an attention-like process was responsible
Downshift unblocking
Just like standard blocking except that two shocks are given in stage 1 with an 8 second gap between them
This stopped blocking from happening
Dickinson, Hall & Mackintosh’s experiment can’t
easily be explained by RWM.
We could modify λ for stage-1. Let’s call it 2λ, to account for the fact that there are 2 shocks. Fine.
When move to stage-2:
∆V1 = αL * β(λ - ΣV)
∆V1 = αL * β(λ - [ΣVL + ΣVT])
∆V1 = αL * β(λ - [0 + 2λ])
∆V1 = αL * β(λ - 2λ)
∆V1 = αL * β(-λ)
- This means that ∆V1 will be negative. So RWM wrongly predicts that that the light will be an inhibitor, not an excitor.
Overshadowing
Stage 1 tone + light , tone overshadowed learning about light
Trial 1 is normal
Trial 2 EV bigger than it would be with just the light, as both the tone and the light contribute to the error term
to RWM overshadowing is just a different way of creating blocking - L and T block one another, if aL = aT learning will stop when EVL = lambdha/2
Who first reported overshadowing
Pavlov 1927
CS-UCS contingency
Contingency theory proposes that for learning to take place, a stimulus must provide the subject information about the likelihood that certain events will occur
Latent Inhibition
Pre exposure to the CS slows the rate of conditioning.
It will get to the same lambdha as it does in non-exposed group, just more slowly
So this is as though the pre-exposure reduces a, it’s just that the RWM has no means of accommodating that.
Phenomena that SOP fixes
One-trial overshadowing (e.g James & Wagner, 1980)
Latent inhibition & its “context specificity”
Phenomena that SOP will FAIL at
Downshifting unblocking (e.g Dickinson, Hall & Mackintosh, 1976)
Phenomena that the Rescorla & Wagner model explains
Conditioning
Configural learning
CS-UCS Contingency
Extinction
Blocking
Overshadowing
Pavlovian inhibition
CCCE BOP
Phenomena that SOP explains
Drug tolerance
One-trial overshadowing
Recognition memory
Latent inhibition
Context-specific latent inhibition
Habituation
DORLCH
SOP stands for
Standard Operation Procedure
Why is SOP different
Ralph Miller described RWM as ‘trial-wise’: We computed change in V after each trial
SOP it operates dynamically in real time
This maps on to the idea that it’s the CS-UCS that’s important
Konorski Psychological representations
Composed of divisible elements (aka. units, nodes, gnostic units)
Explanation of stimulus generalisation
if two stimuli are “similar” is means that they share representational elements e.g Ax and Bx are similar but A and B are not
Hebb suggested…
often summarised as ‘what fires together, wires together’
If they get turned on together, they get connected together
0 < Acs < 1
0 < Aucs < 1
Hebb (1949)
Change in V = pi * (Acs * Aucs)
Hebb can or cannot explain any of the things that RWM can
Cannot
What would be a sensible value for λ when the UCS is NOT presented; e.g., during ‘extinction’ of Pavlovian conditioning?
0
What would be a sensible value for λ when the UCS is presented; e.g., in a normal Pavlovian conditioning experiment.
1
Imagine a Pavlovian conditioning experiment with the first 12 trials having a mild shock, then the next 12 trials having a medium shock. What values of λ could you use?
1 then 2
RWM successfully explains
Blocking
Overshadowing
CS-UCS Contingency effects
Relative validity
RWM fails to explain
“Downshift” unblocking
One-trial overshadowing
Latent inhibition
How RWM fails to explain downshift unblocking
We could modify λ for stage-1. Let’s call it 2λ, to
account for the fact that there are 2 shocks. Fine.
When move to stage-2:
∆V1 = αL * β(λ - ΣV)
∆V1 = αL * β(λ - [ΣVL + ΣVT])
∆V1 = αL * β(λ - [0 + 2λ])
∆V1 = αL * β(λ - 2λ)
∆V1 = αL * β(-λ)
This means that ∆V1 will be negative. So RWM
wrongly predicts that that the light will be an
inhibitor, not an excitor.
Why RWM fails to explain latent inhibition
The pre-exposure to the CS slows the rate of
conditioning. It will get to the same λ as it does in a non-pre-exposed group, just more slowly.
So this is as though the pre-exposure reduces α,
it’s just that the RWM has no means of accommodating that
A1 state
Vivid, perception.
Limited capacity
A2 state ‘primed’
Weaker. Memory trace or activation by another stimulus
Unlimited capacity
I state
Something you’re not sensing or thinking about
Unlimited capacity
How SOP explains one-trial overshadowing
When two CSs (T & L) are paired, each one is less active than they would be alone
this means that less will be learned about them—here that an association between T and the UCS is worse when L accompanies T than if T is there on its own.
So the limited capacity feature of A1 activity literally creates overshadowing of T
SOP’s learning rules
A1 + A1 = increase in excitatory association
CS –> US
A1 (CS) + A2 (US) = Increase in inhibitory association
The representational cycle
An inactive stimulus’ elements (I), become fully active (A1) when is presented, and fade (A2), eventually returning to I
A1 –> A2 –> Inactive
Retrieval-generated priming
Inactive —> A2
An associatively-activated representation goes into from inactive, but only to A2 state NOT A1
At λ the presentation of the CS will “prime” most (but usually not all) of the UCS’s
representational elements into their A2 states.—This is what generates the CR in
Pavlovian conditionin
Process of Retrieval-generated priming
After the first CS-UCS pairing, the CS’s presentation will prime some of the UCS’s elements into their A2 states, producing inhibition
US is presented, elements that didn’t get primed into A2 will be available to go
into their A1 states, producing excitation.
Early in training, few UCS elements will be primed; but as excitation builds up, priming builds up.
At λ, there is no overall change in the balance of excitation and inhibition. Any overshoot of
excitation will be compensated by extra inhibition on the next trial and vice versa.
SOP explains Latent inhibition
the context will retrieval-generate priming in
the CSs representation
This means that the CS’s elements will be in their A2 states, to some extent; ∴ during
Conditioning, they won’t be able to become associated with the UCS.
