T.4: Rescorla-Wagner Model Flashcards
Define V
The associative strength between CS and US
The amount you know
What does a higher value of V mean?
The stronger the associative learning between CS-US
Define delta V
Change in V - how much has been learnt in each trial
What is the R-W model equation?
alpha x beta x (lamda - sigmaV) = deltaV
Define alpha
Salience of the CS
Define beta
Salience of the US
is always valued as 1 in this course
What are the salience factors of the equation?
alpha x beta
What represents the prediction error term in the equation?
(lamda - sigmaV)
What is the prediction error term quantifying?
The notion of surprise
What is the prediction error term?
The discrepancy between the expectation and what actually happens in the trial
Define lamda
Maximum amount of learning that can occur
Magnitude of the outcome
Define sigmaV
Expectancy of the CS-US pairing
What are the 3 steps used when applying this model to a trial of learning?
Step 1: calculate sigmaV before the pairing
Step 2: figuring out how much learning occurs when CS is paired with the US - work out deltaV
Step 3: update the value of V - add original V to deltaV to work out updated V
What happens to the value of deltaV with ongoing trials?
It decreases because learning decreases as the animal is becoming less surprised
The error term has become smaller
What effect does higher salience have on learning?
Increases the rate of learning
What happens to V across trials?
Approaches 1 - expectation of the outcome is being learned
The discrepancy is getting smally
What does delta V approach across trials?
Approaches 0 as the outcome becomes less surprising to the animal
Less learning happening on each progressive trial
What happens to V when you apply the R-W model to extinction?
V gradually approaches 0 - learning is decreasing and losing its associative strength
What happens when a CS is more salient during extinction trials?
The learning decreases faster
Rate of extinction increases
What does deltaV approach when the R-W model is applied to extinction?
Approaches 0
What happens to V when you apply the R-W model to over expectation?
V decreases from 1 to a new asymptote - this is inhibitory learning
What happens to deltaV when you apply the R-W model to over expectation?
The sign (+ or-) of the error term dictates the direction of learning (not just the presence vs absence of the US)
DeltaV approaches 0 on graph example in the slides?
