L9 - Models of Associative Learning Flashcards
Signal validity
We get good learning if the CS is a valid predictor of the US and over and above the presence of the others.
Wagner et al:
Signal validity, shift away from behaviourist thinking
Wagner Experiment to show relative signal validity
3 conditions
- Tone + Light = US
- Tone + Light and just light, both = US
- T + L and Light + nothing both = US
- 2nd result very similar to blocking weak conditioning
- 3rd condition the best response as it shows clearly T is needed for US, light on its own shows nothing
Issues with experiment: animals may habituate, not perfectly controlled for
Wagner Experiment 2
2 groups:
- Correlated (T1 + L = US) (T2 + L = nothing)
- Uncorrelated (T1 and T2 + L both lead to US 50% of the time)
Both groups are given an equal number of L- US pairings (50%) but different in terms of signal validity of accompanying tones
Results:
- Light you don’t learn much about in condition 1, in condition 2 you learn lots
- Condition 1 you don’t as it is being blocked by T1
- Condition 2 - reason for good condition to the light is because it is present more often so therefore it will be relatively speaking a more valid signal
Mackintosh
Variations in conditioning reflect variations in attention to CS
Animals have limited attention resources: must be shared, we can’t take everything from our env in, we have selective attention
Mackintosh Experiment
Conditioned suppression to light
3 conditions:
- Light + US
- Light + noise
- Light + Loud Noise
Variations in attention will occur because the noise in the 3rd condition is very salient, so will capture our attention a lot, suppression ratio closer to 0.5 for light (less learning to light)
Other than intensity what else is attention to CS determined by?
Importance, attention decreases if the animal learns there is a better predictor of US
Blocking is an example of this, first CS is seen as more important.
Wagner Importance Experiment
3 Conditions:
- T + L - US
- T + L - US, L - US
- T + L - US, L - Nothing
Conditioning to T determined by attention devoted to L, condition 2 had a lot of attention to L, condition 3 had little so most conditioning to T
Kamin
US must be surprising, animals won’t learn anything on a trial in which all 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
Surprising theory explaining blocking experiment
Nothing is learned in a blocking design because US is fully anticipated (as signally by pre-trained CS)
Condition:
- N - US, then N+L - US
○ Because noise fully predicts the shock the US is not unsurprising when the N+L leads to shock
However when just N+L leads to shock without prior learning this is a shock so more is learnt
What theory does the Rescorla Wagner Model back?
Model basically an equation capturing the idea that learning occurs when something surprises us
(Kamin’s theory)
The Rescorla Wagner Model
Describes when animals learn associations between CS and US and how fast learning occurs.
Learning is determined by the discrepancy between the experience of the US and how much it was expected (prediction error)
In the RWM the amount of learning equals
how much the strength of the CS-US association changes