5.1 Models of Associative Learning Flashcards
What were the 3 very important associative learning findings published in the same year?
Kamin: blocking
Rescorla: contingency demonstration with truly random design
Wagner: signal validity
What did the 3 findings lead to?
A paradigm shift in how we thought about conditioning
Previously, had been dominated by S-R theory and behaviourists
Now, it was important to understand how conditioning actually occurs
When do we get good learning?
When a CS is a valid predictor of the US over and above the other CS’s present
What is the basic idea behind signal validity?
Learning which is the best predictor of the aversive stimulus
Describe the 3 groups in 1st experiment on signal validity (before it was better controlled)
3 groups association with the tone is equated
Each group has a TL pairing and followed by a shock
Group 1: T+L –> US
similar to overshadowing
Group 2: T+L –> US, L–> US
In addition, intermixed throughout conditioning, you will get light paired with shock as well as TL + shock
Similar to blocking
Group 3: T+L –> US, L–> nothing
In addition, intermixed throughout conditioning, you will get light paired with nothing as well as TL + shock
What was the signal validity experiment observing?
The conditioned response to the tone and to see how the animals differed in their response depending on the conditioning phase
What were the results for group 2 (blocking)?
Similar to blocking:
Weak conditioning
Teaches that the light is the valid predictor of the shock, not the tone
Tone doesn’t add much in terms of signal validity in this case
What were the results for group 3 (L–> nothing) ?
Very good conditioning to the tone
Learn that light is a poor signal of US because it is paired with nothing
Tone is learned as a valid signal predictor of the shock
What was wrong with Wagner’s first experiment on signal validity?
It was not perfectly controlled - the animals may habituate to the shocks
What did Wagner change in the second experiment to make it better controlled?
Used correlated and uncorrelated groups
Everything was made the same other than what was being investigated
T1 and T2 were diff pitches so the animals could discriminate
Describe the correlated group
T1 + L–> US
T2 + L –> nothing
Light is sometimes followed by shock, sometimes not
Tones are correlated with the outcome:
T1 = shock, T2 = nothing
Describe the uncorrelated group
T1 + L –> US/nothing (50/50 chance)
T2 + L –> US/nothing (50/50)
Tones are not correlated with either outcome - there is a 50% chance of getting US or nothing
On testing all 3 CS’s (T1, T2, L), what was found for the correlated group?
High conditioning to T1 and low to T2
Supports signal validity
On testing all 3 CS’s (T1, T2, L), what was found for the uncorrelated group?
T1 and T2 found to have little conditioning occur
L has the highest result as being the best signal predictor
Why is L the best signal predictor in the uncorrelated group?
This is the key demonstration of signal validity:
The light is present more often when the shock is presented and this is what makes it stand out and, relatively speaking, makes it become a more valid signal of the shock
Can still have learning to a cue when it is partially reinforced - this is what is going on with the light
It is partially reinforced greater than the other tones and this is why it is learned better
What are the 2 classes of theories that can explain variation in attention to CS?
Mackintosh argument: conditioning can be explained due to diffs in the amount of attention that is allocated to the CS
Kamin: learning being modulated by how the US is processed (Rescorla-Wagner model)
What was the Mackintosh argument?
Variations in conditioning reflect variations in attention to CS
Attention is a limited resource
We need to be able to allocate our attention towards specific events in the environment
In doing so, it takes away from other things we can give our attention to in the environment
We get conditioning because we attend to certain things in the environment = selective attention
How does attention variation explain overshadowing?
Variations occur when a CS is v salient e.g. loud noise captures our attention a lot - v intuitive
Don’t get much learning to the light that is paired with it because of overshadowing
When something is more salient, we pay more attention to it and so learn more readily about the cue
How does attention variation explain signal validity?
More important signals capture attention better than less important signals
Will give more attention to the valid predictor and devote less attention to the ones that are bad signal predictors
How does attention variation explain blocking?
Attention is allocated to salient, good predictors - use this to explain blocking:
Throughout the pre-conditioning phase, learnt that noise is a v good predictor of US
As a result, this increases your attention to the noise
2nd conditioning phase - because you have learnt that the noise is a good predictor, its (signal validity) salience has increased relative to the light
Focusing attention on noise over light - no attention paid to the light even if equally matched in physical salience
Why do variations in conditioning occur according to Kamin?
Learning is being modulated by how the US is processed - animals won’t learn anything on a trial in which all the events are fully expected
Learning is a process by which we revise what we know about the external world whenever our expectations differ from what actually happens
How did Kamin explain his blocking results?
Explained in terms of surprise = you learn about something if it is surprising
Animal is forming a mental model of what happens in the world and when something is surprising and conflicts with your mental model, this is when you update the model and change your beliefs
Something occurs you don’t expect is when you update your representation of the world - it is to do with the processing of the US
How has the notion of surprise been formalised into a simple model?
The Rescorla-Wagner model
It is a numerical equation capturing Kamin’s idea
What does the R-W model describe?
Describes how animals learn associations between CS and US and how quickly the link is formed
How is learning determined in the R-W model?
Learning is determined by the discrepancy between the experience of the US and how much it was expected
What is the discrepancy in learning referred to in the R-W model?
The prediction error: (lamda - sigmaV)
What is the difference between +ve prediction error and -ve prediction error?
+ve prediction error = when an event isn’t predicted by you and so learning occurs
-ve prediction error = over expecting something and you still get learning but it is reduced
Define V
The amount you know
Everything you might predict
Define lamda
The maximum amount of learning that is supported by the US
Define sigma V
All that you know about all the CS’s present on that trial
How much association is there between the light and US as well as tone and US for example
Define alpha
Salience of the CS
Define beta
Salience of the US
Define delta V
Change in V
i.e. change in the amount of learning
