Blocking

The phenomenon of blocking tells us that what happens to one CS depends not only on its relationship to the US but also on the strength of other CSs that might be present

Kamin (1969) CER paradigm

Group Phase 1 Phase 2 Phase 3

Exp. Gp.

Con Gp.

N Shock

Nothing

NL Shock

NL Shock

L

L

0

.1

.2

.3

.4

.5

Results (CR to L)

Supp.ratio

Exp Con

Learning to the added cue (L) is blocked by prior conditioning with the NSee little CR to L in Experimental group because the N is already a good predictor of the US

The concept of surprise is important in explaining theBlocking effect

Conditioning only occurs if the animal is surprised by the US

T US

On the first trial, the animal doesn’t expect the USThe US is surprising and learning takes place

Over trials the animal learns that the T predictsthe US

The US is no longer surprising and the rate of learningslows down

Typical Learning Curve

2

1

3

4

5

6

1 2 4 5 6 7 8 9 10 11 123

asymptote

On compound trials, giveLT US

Get little or no conditioning to the new L because the USis not surprising; it is predicted by the T

The L is redundant; it provides no new information

What would happen if you changed the US on compound trials?

The US would now be surprising, so should seeconditioning to the added cue

Unblocking

Unblocking is the elimination of the blocking effect – seelearning to the new cue

Holland demonstrated unblocking by changing the US(increased and decreased the US in separate groups)

This experiment demonstrates the importance of surprise

4 groups of rats

Phase 1

L 1 pelletL 1 pellet

L 3 pelletsL 3 pellets

Phase 2

LN 1 pelletLN 3 pellets

LN 1 pelletLN 3 pellets

(1-1) = no surprise

(1-3) = surprise

(3-1) = surprise(3-3) = no surprise

2 groups received the same # of pellets in phases 1 and 2 (blocking)2 groups received a change in the number of pellets(unblocking)

Results

% CRto N

Test sessions

(1-3)(3-1)

(3-3)(1-1)

Unblocking(see learningTo N)

Blocking(no learningTo N)

The Rescorla-Wagner Model

• Based on the concept of surprise• better than simple contingency theory

Contingency theoryAssociations develop when a subject assesses the correlationor predictive relationship between the CS and US

Excitatory conditioning occurs when:p(US/CS) > p(US/no CS)

Problem with contingency theory is that it doesn’t take into consideration what is happening to other CSs; therefore, it can’t explain blocking

The Rescorla-Wagner Model

• US processing model• conditioning depends on the degree to which the US is processed• according to the model, each US has a certain amount of associative strength that will support conditioning

• takes into account all the CSs present on a given trial

• the concept of surprise is important to the RW model•surprise is defined as the discrepancy between the US that is expected and the one that actually occurs

• mathematical model

The Rescorla-Wagner ModelShould be able to grasp the general idea of the RW model if you understand the following 6 basic rules:

1. If the strength of the actual US is greater than the strength of the subject’s expectation, all CSs that are paired with the US will receive excitatory conditioning

2. If the strength of the actual US is less than the strengthof the subject’s expectation, all CSs that are paired with the US will receive some inhibitory conditioning

3. If the strength of the actual US is equal to the strengthof the subject’s expectation, there will be no conditioning

The Rescorla-Wagner Model

4. The larger the discrepancy between the strength ofthe expectation and the strength of the US, the greater will be the conditioning (either excitatory or inhibitory)that occurs

5. More salient CSs will condition faster than less salientCSs

6. If 2 or more CSs are presented together, the subject’sexpectation will be equal to their total strength (with excitatory and inhibitory stimuli tending to cancel eachother out)

AcquisitionL 1 pellet of food

Trial 1:Rat has no expectation of the USSo, the strength of the US is much greater than the rat’sexpectation (which is ‘0’)Therefore, this trial produces some excitatory conditioning (refer to Rule #1)But conditioning is rarely complete after 1 trialTrial 2:The second time the L is presented, it will elicit someexpectation of the US, but still not as strong as the actualUS So rule #1 applies again and more excitatory conditioningdevelops – and so on for trials 3, 4, 5……

AcquisitionL 1 pellet of food

At each conditioning trial, the rat’s expectation of the foodpellet should get stronger

The difference between the strength of the expectation andthe strength of the US gets smaller

The fastest growth in conditioning occurs on the first fewtrials and there is less and less conditioning as the trialsproceed (rule #4)

Eventually, the L elicits an expectation of 1 pellet and 1pellet is given – the asymptote of learning is reached andno further conditioning occurs

2

1

3

4

5

6

1 2 4 5 6 7 8 9 10 11 123

Learning Curve

The RW model predicts the typical learning curve (acquisition)

Blocking

Now, suppose asymptote is reached and we give:

LT 1 pellet of food

According to rule #6, when 2 CSs are presented the subject’s expectation is based on the total expectation from the 2 CSs

T is a new stimulus = ‘0’ expectationL produces expectation of 1 pellet

The actual US = 1 pellet; expectation matches the US that is given and no additional conditioning occurs (rule #3)

The L retains its excitatory strength and the T retains its ‘0’ strength

The Rescorla-Wagner Model

ΔV = k(λ – V)Δ = change so ΔV = change in the strength of the CS

k = constant Related to the salience of the CS and USRefers to the associability of the CS

λ = maximum amount of conditioning that theUS can support – its our actual US value

λ – V = the discrepancy between what the animalexpects (V) and the actual US that is given (λ)

The Rescorla-Wagner Model

Will sometimes see the formula written as:

ΔVA = k(λ – VT)

ΔVA = change in the strength of CSA

VT = strength of all CSs on a given trial(VT = VA + VB ….. )

ΔVA = k(λ – VT)

Pair CSA Food for 5 trials

VT = VA; since only 1 CS

k = 0.5; constant

λ = 100; USTrial 1: ΔVA = k(λ – VT)

ΔVA = 0.5 (100 – 0) = 50

So, change in strength of CS is 50 units

Trial 2: ΔVA = 0.5 (100 – 50) = 0.5 (50) = 25

CS gains additional 25 units

Trial 3: ΔVA = 0.5 (100 – 75) = 0.5 (25) = 12.5

ΔVA = k(λ – VT)

Trial 4: ΔVA = 0.5 (100 – 87.5) = 0.5 (12.5) = 6.3

Trial 5: ΔVA = 0.5 (100 – 93.8) = 0.5 (6.2) = 3.1

ΔVA across 5 trials: 50 + 25 + 12.5 + 6.3 + 3.1 = 96.9

With more trials, V = λ = 100

Most conditioning occurs on trial 1 (50 units)

On subsequent trials the CS acquires additional strengthwhich is a fixed proportion of the strength that is stillavailable

As conditioning progresses, the discrepancy between theexpected and actual US declines; (λ – VT) gets smaller

So, the RW model predicts the typical learning curve

20

40

60

80

100

1 2 3 4 5 6Trials

VCS

or CR

Learning Curve

Rescorla-Wagner Model and Blocking

At the end of trial 5, in the previous example, the total strength of all CSs is 96.9 (VT = VA = 96.9)

This means 3.1 units from the original 100 are availableon trial 6

Trial 6: CSA/CSB Food ΔV = k(λ – VT)Assume k = .5VT = VA + VB

ΔVB = .5 (100-96.9)= .5 (3.1)= 1.5

Recall, CSA on first trial gained 50 units of strength

But, CSB gained only 1.5 units of strength

Conditioning to CSB is blocked

The RW model can explain acquisition and blocking

The RW model can also explain extinction andconditioned inhibition

ΔVA = k(λ – VT)

Suppose after asymptote is reached with L-Food pairingswe give: LT No food Assume k = .5

VA = 100 (asymptote)Here, rule #2 applies:

The strength of the expectation will exceed the strength of the actual US

λ = 0 (no US is given); VT = 100, because L is given

According to rule #2, both CSs will acquire some inhibitory strength

How does this inhibitory conditioning affect the L and T?

Because the L starts with strong excitatory strength, the trials without food and the inhibitory conditioning they produce will begin to counteract this excitatory strength

This is an example of extinction – presenting the CS without the US

ΔVA = k(λ – VT) ΔVA = .5(0-100)

= -50L starts with 100, now reduced to 50 – but still excitatory

Repeated trials without the US would reduce L to ‘0’

The T begins this phase with ‘0’ strength because it hasn’t been presented before

Therefore, trials without food will cause the strength of the T to decrease below 0 – it will become a CI

Remember, L food (until asymptote)Then, LT no food

ΔVA = k(λ – VT) ΔVA = .5(0-100)

= -50

T starts with 0, now at –50; so its inhibitory

The RW model can also explain the US-pre-exposure effect

On US alone trials, the background cues become conditioned

Phase 1: US alone

Phase 2: CS US pairings

Then when CS-US trials are given, it becomes a blocking experiment

Context + CS US

Unusual prediction of the RW Model

Loss of associative value despite pairings with the US

Phase 1:

A USB US

Phase 2:

A + B US

A and B are trained to asymptote in phase 1

Then in phase 2, both CSs are presented together withthe same US

The expectation in phase 2, would be 2X the US

However, only 1 US is given, so the expectation exceedsthe actual US

So, should see a decrease in CR to both A and B

Evaluation of the RW Model

The RW model cannot account for latent inhibition (LI)Phase 1: CS alone

Phase 2: CS US pairings

According to the model the CS should not gain or lose strength when no US is present

The RW Model is a US-processing model

ΔVA = k(λ – VT)

ΔVA = .5(0 – 0)= 0

So, change in strength of CS on first trial is ‘0’

The next trial would be the same

No increase or decrease in strengthof the CS

Evaluation of the RW Model

Another problem for the model is unblocking

There are 2 types of unblocking:

Unblocking with an upshift (i.e., the US is increased)

Unblocking with an downshift (i.e., the US is decreased)

The RW model can explain unblocking with an upshift If the US is increased, λ is increased, there is room to seeconditioning to the added CS

The model cannot explain unblocking with a downshift If the US is decreased, λ is decreased. Should never seeexcitatory conditioning to the added CS

Evaluation of the RW Model

ExtinctionAccording to the model, extinction should reduce the strength of the CS to ‘0’i.e., extinction is the reverse of acquisition

However, we know that extinction is the not the reverse of acquisitionSpontaneous Recovery: if we give the animals a restperiod, responding the CS recovers

Temporal factors in conditioning

Temporal factors like the CS-US interval are important but the RW Model cannot account for these factors