Topic 8 Flashcards
principle 3.1
learning is what happens when our predictions are wrong
principle 3.2
everything gets boring eventually
Pavlov and his dogs
- led to the principle of contiguity (the more often two things occurs together over time we will associated them together) during training every time the dog gets food there is a whistle
- results in conditioned stimulus - dogs mouths watering at the sound of the whistle
classical conditioning
- we naturally respond to things in certain ways
- by pairing new cues with that natural stimulus we respond to the new cue in the same way
- learning is defined as the change in strength of a behaviour in response to a previously neutral cue
- change in behaviour indicates a change in association
- leads us to respond to cues, like habits automatically
- partly responsible for tolerance to substances - environment triggers not personal attributes
- is also responsible for emotional responses. we have a habit of how we respond to social situations
acquisition of learning
the first step of classical conditioning
definition: learning to associate a conditioned stimulus with an unconditioned stimulus so that when the conditioned stimulus is presented on its own, it elicits the same response as the unconditioned stimulus
acquisition timing of cues
- delayed conditioning - food right after
- trace conditioning - pause before food
- simultaneous conditioning - cue and food at the same time
- backwards conditioning - food and then queue
extinction definition
weakening of an association between two stimuli over time - unlearning
eg: whistle with no food to make there be no salivation reverting it to a neutral stimulus instead of conditioning
the Rescorla Wagner Model (2 assumptions)
- learning occurs when what happens does not match our predictions (we are wrong)
- our predictions are based on previous experience
the Rescorla Wagner model definition
whenever there are two stimuli (or more) paired together, any change in the strength of the association between two stimuli will be proportional to the difference between what was expected and what happened and how noticeable the association is
the Rescorla Wagner model formula
At=S(W-At)
At= change in strength/association on current trial for stimuli
S= silence (range 0-1) how noticeable 1=instant learning
W= maximum strength of the association
At = current strength of all associations - how much learning has occurred so far
(W-At)= prediction error
learning with the Rescorla Wagner Model
light with food
W= 100 - the amount of reward being paired with the light
no learning yet so dog does not associate light with food therefore
A1= 0
not obvious that light is associated so
S=0.2
A1=0.2(100-0)
A1=20
aka there is an increase in the strength of association
therefore even if light were only presented dog would expect 20 units of food to be deliver
- not 20
the Rescorla Wagner model trial 2
A2= 0.2(100-20_
A2= 16
less than before but we continue to learn
the Rescorla Wagner model trial 3
on trial 2 we add A2 + A2 aka 20 + 16 therefore
A3 = 0.2(100-36)
leading to a smaller change in the strength of association
A3 = 12.8 = 0.2(64)
does the rate of learning slow down after each trial? (RW model)
yes, but from 10 trials u can go from 0% association to 99% association
key notes of Rescorla Wagner model
by trial 10 the change per trial is very small
the current strength of association gets closer and closer to the W
common saying in RW
“that got a steep learning curve”
RW model shows us that it is not always that something is difficult
how are the trials represented in a graph? (RW model)
by a curve that starts at 0 and then increases and increases and eventually starts to plate
extinction with RW
light with no food
A1= 100
W=0 - no food
is it immediately obvious that there is no food coming? NO
S= 0.2
A1= 0.2(0-100)
same as the first one but negative
Trial 2
A2= 0.2 (0-80)
A2= -16
continues.. the graph goes downwards
blocking
acquisition phase continues what happens when we add a light
RW for stimulus 2
A2t= s(100-95)
the acquisition has already occurred so the prediction error is low
so how much learning is possible?
5 - split by the two stimuli
graphs of blocking
two lines, one at the top and one at the bottom
A1 is normal and the A2 comes later but doesn’t rise
acquisition prior to over-expectation
the dog learns light and whistles both mean food
because the dog learned each separately there is no blocking
light = 100 units of food
whistle = 100 units of food
graphs are normal and similar
when does over-expectation occur
ex: dog thinks when its presented with both the whistle and the light it will get double the food
the chart starts high and goes lower because the dog is disappointed
(disappointment curve)
overshadowing
what if two stimuli are learned together.. but one is much more noticeable than the other?
salience influences how quickly learning will occur
spontaneous recovery
RW predicts that if response to CS undergoes extinction, it must be relearned from scratch for there to by any response to it again
small response after relearning something
rapid reacquisition
RW predicts learning should take place at the same rate as before, but this is not the case
learning is faster the second time
latent inhibition
once you pair stimulus with a reward, should begin at the rate determined by the salience of the associated in that first trial seeing a stimulus do nothing before makes it harder to associate it with a reward
- decreases salience
limitations of RW
fails to predict some things, like all models, it is wrong in some ways and taken over by more complex models
the point of RW
show how learning depends on errors
demonstrates some common learning phenomena to be aware of
show how learning happens quickly at first but slows over time
habituation
occurs when the response to a stimulus decreases over time after a long time, habituation occurs
- faster on round 2
- higher intensity stimuli lead to higher intensity responses and slower habituation
- overlearning can occur.
continued exposure after habituation can have a long term effect
the good side of habituation
- distractions - things that distract us in the environment can often be tuned out with repeated exposure to them
- overcoming fears: by exposing yourself to things that cause anxiety, you can slowly reduce your anxious response to those things and be more in control around them
the dark side of habituation
too much of a good thing - get less excited about it over time
eg: substances
- compassion fatigue: being exposed to sad stories leading us to be less likely to help
- learned helplessness: overexposure to stressful environment may make us less willing to escape the environment
graded exposure
take a fear, then expose yourself to that fear until slowly over time your less scared
factors that impact habitation
novelty: if the duration or intensity of stimulus changes your response will reset partly
duration: if the duration is too short, habituation to future exposures will be slower
frequency: more frequent exposure leads to quicker habituation
intensity: higher intensity = harder to tune out
the inherent reward for learning
dopamine firing shifts from reward to the stimulus cue - our brains reward us for learning
learning is dependent on prediction error
rewards lose their impact over time RW is steepest in the beginning + habituation
stimulating a task choice
one you learn a hobby you become bored if it is too easy