Download PSY402 Theories of Learning

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Impulsivity wikipedia , lookup

Insufficient justification wikipedia , lookup

Behavior analysis of child development wikipedia , lookup

Verbal Behavior wikipedia , lookup

Behaviorism wikipedia , lookup

Psychological behaviorism wikipedia , lookup

Emotion and memory wikipedia , lookup

Allochiria wikipedia , lookup

Perception of infrasound wikipedia , lookup

Classical conditioning wikipedia , lookup

Psychophysics wikipedia , lookup

Operant conditioning wikipedia , lookup

Transcript
PSY402
Theories of Learning
Chapter 10 – Stimulus Control of
Behavior
The Role of Environmental Stimuli

In operant conditioning, the
stimulus becomes associated with
the reinforcer or punishment.



Reward or punishment is the UCS.
The stimulus signaling reward or
punishment is the CS.
The CR motivates operant behavior.

Responding can be used as a measure
of the strength of a CR.
Definitions of Terms



Stimulus control -- Environmental
stimuli signal the opportunity for
reward or punishment.
Generalization – responding in the
same way to similar stimuli.
Discrimination – responding to
some stimuli but not to others.
Generalization Gradient

Degrees of generalization occur.



In some situations, the same response
occurs to similar stimuli.
In other situations, the amount of
response varies with the similarity.
Generalization gradient – a graph
showing how the strength of
response changes with similarity.

Steep gradients mean narrow response
(stimuli must be very similar).
Kinds of Gradients

Excitatory conditioning (S+) – a CSUCS response to a stimulus is
learned.


Excitatory gradient – the S+ is varied
and the CR is measured.
Inhibitory conditioning (S-) – a CS
signals absence of the UCS and thus
inhibits the CR.

Inhibitory gradient – the S- is varied
and the CR is measured.
Wavelengths of Light
yellow
yellow-orange
yellow-green
green
bluegreen
orangeyellow
orange-red
orange
red
Discrimination


The shape of the gradient can be
changed by training.
When birds are exposed to two
different tones (S+ or S-), they must
discriminate between them.


Responding is less generalized because
the competing tone produces no
reward.
The shape of the gradient becomes
steeper and more narrow at the top.
The sharpness of the generalization gradient
depends on the type of training
Flat Gradients


A flat gradient means all stimuli are
being responded to as if they were
the same.
Responding with a gradient to a
tone occurred only when the tone
signaled reward during training.
Generalization of Inhibition

Inhibition example: fear of dating.


A good experience leads to less fear of
dating a different person.
Inhibition gradients are similar to
excitatory gradients – the more the
stimulus varies, the less inhibition.
Excitatory and Inhibitory
Generalization with Line Tilt Stimuli
Explanation

Lashley-Wade theory – people and
animals generalize because they are
unable to discriminate.




Can’t tell the difference between stimuli
A contrast is needed during training to
enable discrimination.
Discrimination training leads to
steeper generalization gradients.
Perceptual experience matters.
Discrimination Learning


Important to recognize when
reinforcement is not available so
that responding can be withheld.
Discriminative stimulus:



SD – reinforcement is available (S+)
SD – reinforcement is unavailable (S-)
Conditioned stimuli always produce
a response. Discriminative stimuli
signal the opportunity to respond.
Two-Choice Discrimination Tasks

The discriminative stimuli are on the
same dimension:




Red vs green light.
Need not be presented
simultaneously.
Two-choice discrimination includes
one SD and one SD .
Other tasks can use multiple
multiple SD or multiple SD.
Categorization and Discrimination

Animals respond to stimuli in ways
that suggest they form categories.



Pigeons can classify a variety of items,
including new images not seen before.
The items to be learned as members of
a category are SD and signal
opportunity for food.
The items that are not members of the
category are SD and signal that pecking
will not be rewarded.
Test Slides – Tree Category
Test Slides – Water Category
Test Slides -- Margaret Category
More Complex Tasks

Later pigeons were asked to place
images into four categories by
pressing one of four buttons
(rewarded by food if correct).



They are “naming” the object shown.
Pigeons do equally well with natural
and manufactured objects (cars,
chairs).
Transfer to new stimuli is worse but
above chance.
Apparatus (Part 1)
Examples of positive images
Examples of positive images
Three Phases



Subjects begin by responding
equally to both stimuli –
prediscrimination phase.
Discrimination phase -- with
training, response to SD increases
and response to SD declines.
Shift back to non-differential
reinforcement to show that behavior
was caused by reinforcement.
Conditional Discrimination

Availability of reinforcement
depends on the condition of a
stimulus.



The stimulus does not always signal the
same thing.
More difficult to learn.
Nissen’s chimpanzees:


Large, small squares, white or black.
SD = large white, small black.
Behavioral Contrast


Behavioral contrast – the increased
responding to the differential
stimulus, decreased response to SD
Contrast also occurs with changes
in the duration of reinforcement.



VI-10 to VI-3
Local contrast – emotional
Sustained contrast – related to the
differential reinforcement.
Anticipatory Contrast

Williams – sustained contrast occurs
due to anticipation of a future
reinforcement contingency.


Not due to recall of past contingency.
VI-3, VI-6, VI-3


Compare the first and third VI-3
behavior.
VI-6 affected the first VI-3 more than
the last VI-3 -- a prospective effect.
Occasion Setting


A conditioned stimulus (CS) can
create the conditions for operant
responding to a second conditioned
stimulus (CS).
Occasion setting – ability of one
stimulus to enhance the response to
another stimulus.

The facilitating stimulus does not
produce a CR by itself.
SD as an Occasion Setter


A Pavlovian occasion-setter can
increase operant responding.
Example:



A meal elicits CR craving for cigarette.
Requesting a cigarette after a meal –
an operant behavior caused by CR.
Conditional occasion-setting:

Second stimulus modifies meaning of
first discriminative stimulus.
Conclusions



An occasion-setter can increase
operant responding.
A discriminative stimulus (SD) can
increase response to a CR
(Pavlovian conditioning).
This implies interchangeability of
Pavlovian occasion-setters and
discriminative stimuli.
Peak Shift

When both inhibitory and excitatory
stimuli are conditioned, inhibition
changes the shape of the gradient.



Peak shift – maximum responding
occurs to a stimulus not previously
trained as the S+.
The peak shifts away from the Sstimulus.
The amount of response is the
difference between inhibitory and
excitatory conditioning.
Hypothetical Excitatory and
Inhibitory gradients
Spence subtracts the
inhibition on the next
slide from this excitation
Hypothetical Excitatory and
Inhibitory Gradients
Overall predicted
response is less because
this amount of inhibition is
subtracted from it.
Peak Shift
When the inhibitory
stimulus S- is to the
right, the peak shifts
left
Errorless Discrimination Learning


When an SD is gradually introduced
the pigeon learns to inhibit
response without making mistakes.
Three fading steps are involved:



Brief introduction of SD for 5 sec-30 sec
Slowly change color of SD from dark to
green
Slowly increase duration of SD from 30
sec to 3 minutes
Implications of Errorless Training

Errorless learning seems to
condition response to SD without
inhibition to SD.



This means that errorless learning is
not aversive.
As a result, no peak shift occurs.
Errorless learning is harder to
condition to some stimuli than
others (e.g., colors but not lines).
Application of Errorless Training

Examples with humans:




Preschool children recognizing shapes
using a fading technique.
Oral reading.
Dorry & Zeaman taught mentally
handicapped children to identify
vocabulary words (pictures faded out).
Not all training works – problems
with transfer and with reversed
consequences.
Is Learning Relational?


Are animals learning the
relationships between stimuli rather
than an absolute response?
Transposition occurs when stimuli
are changed:


The brighter of two lights, louder of two
tones is responded to.
Different results support both views
of learning: Hull-Spence & Kohler.
Mackintosh’s Attentional View

Stimuli with multiple dimensions
arouse the relevant dimension
analyzer.



This depends on the salience and
intensity of the dimension.
The predictive value of the dimension
determines arousal.
Discrimination learning depends on
predictiveness.
8.17 Examples of computer stimuli presented to pigeons by Cook (Part 1)
8.17 Examples of computer stimuli presented to pigeons by Cook (Part 2)
8.17 Examples of computer stimuli presented to pigeons by Cook (Part 3)
Less popout with conjoined
features.
8.18 “Same” and “different” displays used in the experiment by Wasserman et al
Continuity Theory

Hull-Spence suggest that excitation
and inhibition gradually increase
with trials.


Non-continuity theory suggests that
a hypothesis is formed & tested.


Excitation to SD, inhibition to SD.
Learning occurs rapidly with attention
to the right dimension.
There is support for both theories.