Learning Flashcards
why is it important to learn
to make predictions and control events in the future
learning helps organisms to benefit from the natural events through the nature’s regularities
What are the two factors to examine the causal effect between two events
Degree of covariation
> whether two events happen together or independently
Degree of contingency
> If so, to what extent do they co-occur
What is Delta(P), what is the formula, and what are the findings of Shank’s diagnosis study and Chatlosh’s study
Delta(P) = to measure the actual strength of contingency between a cue (vaccine) and an outcome (covid)
formula = P[A|B] - P[A|not B]
delta(P) closer to 1 = stronger positive relationship
delta (P) closer to -1 = stronger negative relationship
delta (P) closer to 0 = weaker relationship
Shank’s study
> people can predict the direction of the Delta(P)
> However, sometimes over-estimating it
Chatlosh’s study
> The number of blocks showing the events is positively correlated to the accuracy of the prediction of the Delta(P)
Explain when should we use the probabilistic contrast model, and what is the takeaway of Chapman & Robin’s experiment
only when the background is constant that we can use delta(P)
takeaway:
sometimes the Delta(P) is inaccurate, if the background noise is significant
humans are capable of perceiving the Delta(P) correctly without the hindrance of background noises
What are the two conclusions that can be drawn from the blocking paradigm
- Learning is not just merely a reflection of co-occurrence. If it was just a reflection of co-occurrence, it would have been a 50:50 chance of getting food or drink
- The early pairing of the CS is more influential than the later CS
Explain the delta rule, the components, the formula, and the learning rate
Delta rule can explain classical conditioning
lambda (strength of the US), Vx (the strength of the CS), gamma (the learning rate)
formula: new Vx = old Vx + delta Vx (gamma x (lambda - old Vx))
learning rate: rapid learning rate in the beginning, slows down and eventually flattens in the end.
Explain how can the Delta Rule explain blocking effect, and the relationship between Delta Rule and contingency learning
Delta rule emphasises on learning through reward prediction error (the higher the error is = the more significant and faster learning is)
Reward prediction error = actual reward - expected reward
the initial stimulus = learn quickly and significantly since the reward prediction error is high
however, when the second stimulus is presented, the reward prediction error is already reduced > therefore the learning effect = not as significant as the first stimulus
The higher the Delta(P) (i.e., contingent) is = the better the prediction
What are the two problems that Delta Rule fails to explain
Highlighting effect
> red light + bell = food; later showing red light + alarm = juice
> when bell + alarm = juice
In other words, the learning effect of the two items are the same, whilst the first is expected to be more significant
Quick learning
> with the aid of the blicket detector
> kids can learn quicker than than the Delta Rule predicts
Retrospection Revaluation
> Backward blocking:
two blickets > one blicket produces noise
> Indirect screening off
two blickets > one blicket doesn’t produce noise
Under both conditions children can accurate predict the outcome according to Delta(P)
How can selective attention explain highlighting and blocking effect
highlighting effect
red + bell = food
later showing red + alarm = juice
since alarm and juice = unique items that have never occurred before
therefore > attention shifts to alarm and juice
when show bell and alarm = juice
blocking effect
red = food
later showing bell = food
the reward prediction error = already reduced therefore effect of learning is smaller than blue and alarm = juice that are introduced at the same time
therefore, when alarm + bell = juice
what does the experiment on filtration and condensation tell us about
filtration condition = learnt quicker as there is only one dimension that differs
condensation condition = learnt slower as there are two dimensions to focus
however, the delta rule does not capture the difference in learning rate. Delta rule predicts the learning curve for both conditions overlap
but with selective attention added to the delta rule = successfully captured
Explain the concept of belief updating (Bayes rule, the mechanism, what can it explain and fail to explain, the two components)
Bayes rule: update weights based on balancing evidence
the mechanism: learn through balancing possessed beliefs and data (possessed belief = the belief that the person owns; data = the experiences in the world)
if the person does not have strong beliefs at the beginning, learning = driven by the data
if the person starts with strong beliefs = need a lot of data tp learn
Two components:
holmesian deduction: when everything that is impossible is eliminated, anything that is left must be the truth
judicial exoneration: if one hypothesis is explained by data, the other candidates are seen to be less likely
Explains the blocking effect and retrospective revaluation. However, if selected attention is added = can also explain highlighting effect
