Rescorla & Wagner model Flashcards
What is the relationship in surprisingness of US and learning ?
A US can be surprisingly big = excitatory effect > increase in associative strength
US can be surprisingly small =
Inhibitory effect > decrease in associative strength
what is the effect of US expectation according to R-W?
US expectation is based on the already acquired associative strength of all the stimuli that are present in a certain trial
Strong CR = index of strong expectation US
Weak CR = indication low expectation US
What is λ in R-W model?
λ = asymptote of learning possible with the US
(will be 1 when US is present in trial,
Will be 0 when US is not present in trial)
What is ΔV in R-W model?
ΔV = change in associative strength of a stimulus on that trial
What is V in R-W model?
sum of already required associative strength through all the CS on a trial
What is (λ – V) in R-W model?
how much is the US surprising = discrepancy of expectation
the amount of learning on a trial
= large at start of learning , because associative value V is close to zero
1-0 = 1
what is the formula for 1 CS?
ΔVA = k (λ – VA)
k = salience - constant
how does the aquisition of associative strength change over trials with 1 CS?
beginning: VA = 0, dus (λ – VA) = (1 – 0) = 1 = big (strong underprediction cause low expectation), so ΔVA= big increase
After a few trials: VA = big, so (λ – VA) = small (small underprediction, high expectation ), dus ΔVA = only small increase
Typical acquisition curve = negative growing curve > first it has steep increase, after it the growth becomes less
What is the formula for 2 CS in associative learning?
ΔVA = k(λ – VAB)
ΔVB = k(λ – VAB)
Where > VAB = VA + VB
According to Rescorla-Wagner this implies:
competition between CS-A and CS-B for associative strength with US because
Sum VA + VB = totally λ (and not more)
See overshadowing!
What is overshadowing according to R-W?
After A+ training CR on A will be bigger than after AB+ training
B “overshadows” acquisition of A
B also gains associative strength during AB+ training
overshadowing:
during simultaneous training of CS A and CS B > VA and VB will be smaller then if they were trained independently!
competition
What is Blocking mechanism according to R-W?
Phase 1: CS A is trained until a perfect expectation of US is acquired. At the end of phase 1: VA = λ = 1
Phase 2: Stimulus B is presented together with A and followed by US.
no conditioning will occur for B because US is perfectly predicted by A.
(λ - VA+B)= 0 in phase 2
How does unblocking work according to R-W?
Unblocking > through stronger US on trial
Phase 2: the US is increased 2 times in this phase which means learning space λ is bigger.
2λ in this case. Adding of CS - B will have this effect:
A has association equal to λ = 1
B has association equal to 0 (because no association build yet)
The λ in the formula is double, there exists an underprediction in the second phase, - so there is still space to learn for CS B.
New formula = (2λ - VA+B)
No blocking for B > it becomes excitatory!
How can Loss of associative value ↓ happen - despite pairing with US?
answer: Over - expectation
Phase 1 = A and B are associated with US > but on separate trials!
They both gain associative value > until they both predict US perfectly = λ
Phase 2 = A and B are presented simultaneously on the same trial with the same US as in phase 1.
The subjects expect to get a US strength of A and B combined > but this is not the case.
RESULT: They will have to lower their expectation to meet the real condition > loss of associative value
How does conditioned inhibition work?
Once CS-A has acquired excitatory value and is predicting the US well, it is paired with the CS-B on trails without the US.
This creates overexpectation for CS-A.
To predict absence of US on non-reinforced trials -> the associative value of CS-A + CS-B have to sum to zero. Because CS-A has acquired positive value, CS-B has to than gain negative value for it to sum to 0.
ASSUMPTION = CS- acquires negative associative value
What are the Conditioned inhibition extra facts?
- During A+B no US trials > A will lose some of its excitatory value
- Acquiring positive association CS+A goes faster, than acquiring negative association CS-B
- Underprediction (1-VA) trials is bigger than overprediction op (0 - VAB) trials
Will stabilize when A=λ and B= -λ
