5.2 The Rescorla-Wagner Model Flashcards
Why do we bother having a numerical formula?
If we have a clearly laid out model, it allows us to make predictions
We can design experiments and see if the empirical findings match the predictions of the model
It is testable
What happens to deltaV across trials?
It slows - the change in v is larger in earlier trials compared to later
This is because we learn more about things earlier on
The curve decelerates = Rate of learning plateaus to an asymptote - gets closer to a value of 1
The value of 1 refers to lamda - this is the maximum amount of learning that can be supported by the US
Explain lamda in terms of a pie
The whole pie is all that can be supported by your US
On any given trial, you take a slice of that pie
On 1st trial, you can learn the whole pie but don’t expect anything yet
Learnt a quarter of the pie in the example = learning a 1/4 of all there is to learn about the CS-US relationship
As each trial progresses, you take more slices from the pie and learn more
However, you get less hungry so you take smaller slices of the pie
Still learn at the same rate tho - still eat the pie at the same rate
How do we quantify the no of magazine entries by the rat in a Skinner box?
Every time the rat pokes his head into the magazine, breaks the infra-red beam and quantifies the no of magazine entries
Can think about this as the amount of expectation the rat has about when food is going to be delivered
Give an example of the R-W model in action through the rat in a Skinner box receiving food pellets
Prior to delivery of food, display a light or a noise for 20 secs before food
Across trials, rat should learn that light/noise predicts the delivery of food and therefore, there should be conditioned responding in terms of its expectations that the food is going to arrive
Across days, the learning increases
If we use the R-W model, we get a nice curve that fits the experimental data
Good models allow us to fit real data to a curve
What happens when the CS is more salient?
Alpha increasing results in an increase in the rate of learning
Higher salience CS’s condition faster compared to lower salience CS’s
How does R-W model explain overshadowing?
Both CS’s are equally salient
Simulation shows us that 2 CS’s result in less learning compared to 1 CS presented alone
The 2 CS’s have to share the pie - they share “V”
How does R-W model explain overshadowing?
1 CS is more salient than the other
Red = more salient, blue = less salient, green = 2 equal salience CS’s
Blue learns at a slower rate because presented with something else that is more salient
This is why it is learned slower than the green even though they have the same salience - depends on what they are being presented with
Green is presented with something of equal salience
Blue is overshadowed by the red in its salience i.e. less pie is left for the blue CS
How does the R-W model explain blocking?
Blocking has 2 phases: stage 1 and 2
Stage 1: 1 CS (A) is present and reinforced = pre-training i.e. learn 1/2 of the pie
Stage 2: 2 CS’s (AB) present and reinforced = training
The pre-trained CS blocks learning to the added CS
Has already taken 1/2 the pie - sigma V is already large before the new CS is added
When you add in another CS, there is less of the pie to take and therefore delta V is small on each training trial with the added CS
When you sum up everything, there is good conditioning to the pre-trained CS but v poor to the 2nd CS
because it has a head start and a blocking effect on the added CS
How does the R-W model explain signal validity?
Intermixing the pairings : sometimes TL pairing, sometimes just L
More learning to the light across time as there have been more opportunities to take a slice of that pie
Robs the tone of its fair share of the pie and results in better conditioning to the light
How does the R-W model explain the truly random design?
The context or environment is itself a CS
What happens in the truly random trials, the rats do still learn something but specifically not learning that the CS predicts the US
They are learning that the context is predicting the US - context predicts the occurrence of the shocks
Context is always there in the background
What was the experiment Odling-Smee ran in 1975?
Showed that extra “unsignalled” US presentation caused considerable conditioning to context
What did Odling-Smee use in his experiment and why?
Place preference chamber
Rats like more time in the black chamber because they are nocturnal animals
Can treat this as a context
Design an experiment where we vary the contingency of the tone shock association in the black chamber
Allows us to reveal how much is learned to the tone and how much is learned to the context
What did Odling-Smee do to 6/8 conditions in the experiment?
Varied the proportions of contingency between US and CS
Depending on the proportion results in how many times the CS-US pairing occurs in the black chamber
What happened to the rats when they had high proportion of CS-US pairing?
High proportion of CS-US pairing, learn the tone predicts the CS and move out for the shock but still spend most time in black box
Do not associate the US with the context, associate it with the CS
