PSY260H1 Lecture Notes - Lecture 6: Eric Kandel, Vending Machine, Operant Conditioning Chamber
PSY260H1: Notes continued: OCT 17, 2017
Missed notes (came late)
Rescorla-Wagener Model
1. Each Conditioned stimulus has a weight (W) for predicting a US
So these weights go from 0 (no expectation, neutral stimulus) to 1.0 (certain)
2. The expectation of a US is based on a sum of weights for all the CSs present.
3. When the prediction is wrong, the weights are adjusted.
Learning is proportional to the prediction error.
So the prediction error is the actual US – Expected US
Adjustment: each active stimulus adjusted by a proportion of Error.
Change in stimulus weight = Learning Rate x Prediction Error
When error is 0, no learning occurs (asymptote)
Look at visual graph online (go through all this in the textbook)
Key idea: the amount of change that occurs in the association between a VS and a US depends
on prediction error, the difference between whether the animal expects the US and whether the
US actually occurs.
Other learning graphs on next page (Example: Blocking)
As I understand it, the basic premise is this. What happens minus what you predict is your
prediction error. And in order to consistently be able to predict stuff you need to have more
trials.
CS Modulation Theory:
Mackintosh has proposed a model of classical conditioning focused on attention, and the way the
CS is processed:
Stimuli have a salience that determines attention.
Repeated exposure with no consequences decreases salience (attention), a form of habituation.
In latent inhibition, pre-exposure to the CS decreases attention for that stimulus, making it harder
to learn about it in the training phase.
So the analogy about Stock Broker An vs Equally amazing Stock bro Steve who is actually just
useless to you because you have An is very similar.
So in the previous (Wagner) model you would be like okay Steve you can come in (you have a
weight of 0 though)
Whereas in this case we don’t care about Steve at all and don’t let him in.
Brain Substrates:
So since Pavlov there has been keen interest in understanding how classical conditioning works
at the neural level. In the 1980s, Thompson and associates discovered that eye-blink conditioning
in rabbits depends on the cerebellum. (Specifically: not the ability to blink but the conditioning).
CS input pathway:
Graph on slide.
Sensory input processed in various brain regions. So all input comes in from the brain stem and
then is going into the cerebellum.
So Sensory input processed in various brain regions (Thalamus, Occipital Lobe, etc.)
Then Sensory info goes to the sensory nuclei of the pontine nuclei.
Mossy Fibers split:
1) To granule cells in Cerebellar Cortex
2) To interpositus nucleus within the cerebellum
US input pathway:
Inferior olive of midbrain to climbing fibers that split:
1) To interpositus nucleus
2) To Purkinje neurons in the cerebellar cortex.
OUTPUT:
CR Output: Purkinje cells of cerebellar cortex, collect both CS and US input.
Inhibits output neurons of the interpositus nucleus which can activate CR. So it is only when you
inhibit the purkinje cell that the nucleus can then fire.
Inhibits output neurons of the interpositus nucleus which can activate CR.
Interpositus output also inhibits intial stage of the US pathway.
Interpositus can generate CRs (e.g. blink to tone), but does not generate URs (e.g., blink to
airpuff)
So why is there no UR in this figure (By this figure we mean the figure on the slides which you
need to refer to).
So the reason for this is that the actual blinking is not a FUNCTION of the cerebellum. For
instance, in the lesion example where they lesioned the rabbits and they could no longer perform
classical conditioning they still could blink. They just didn’t have the learning mechanisms. So
they don’t need the input from both pathways in order to blink but you do need in order for the
conditioned response to occur. (Not super clear refer to slides and textbooks).
CS-US association may be stored in purkinje cells of the cereberall cortex
(Missed notes) –Get from slides
Graphs depicting conditioned and unconditioned stimulus coming (Electrophysiological
Recording in the Cerebellum)
2
PSY260H1 Full Course Notes
Verified Note
2 documents
Document Summary
Rescorla-wagener model: each conditioned stimulus has a weight (w) for predicting a us. So the prediction error is the actual us expected us. Adjustment: each active stimulus adjusted by a proportion of error. Change in stimulus weight = learning rate x prediction error. When error is 0, no learning occurs (asymptote) Look at visual graph online (go through all this in the textbook) Key idea: the amount of change that occurs in the association between a vs and a us depends on prediction error, the difference between whether the animal expects the us and whether the. Other learning graphs on next page (example: blocking) As i understand it, the basic premise is this. What happens minus what you predict is your prediction error. And in order to consistently be able to predict stuff you need to have more trials. Mackintosh has proposed a model of classical conditioning focused on attention, and the way the.
