1
|
|
2
|
|
3
|
|
4
|
- Goal – understand how behavior comes under stimulus control.=
li>
- Solution – treat the pigeon as a psychophysical observer. Sign=
al
detection theory provided an excellent framework for our initial
analyses of the choice behavior of our pigeon subjects.
- Fundamental assumption – discrimination training is the basic
process through which learning occurs.
|
5
|
- If the task is difficult there is an initial period during which there is no evidence of the
discrimination. This period of chance accuracy is referred to as the=
presolution
period (PSP).
- As training progresses there is a gradual increase in acc=
uracy.
- The length of the PSP and the asymptotic accuracy is a function of t=
he
difficulty of the discrimination.
|
6
|
- It is assumed that the sensory effects produced by the stimuli vary
normally.
- A criterion is set on the sensory effect axis dividing the
continuum two categori=
es,
here R1 & R2.
- New stimuli are categorized accordingly. The proportion of R2=
sub>
responses approximates a normal ogive as shown in B
|
7
|
|
8
|
- In this illustration a soft sound when paired with a dim light belon=
ged
to one category and a loud sound paired with a bright light belonged=
to
the other category.
- Pigeons (and humans) use both stimulus dimensions in deciding between
the two categories.
- Decisions based on a line cutting across the bivariate distributions=
are
more accurate than for either dimension alone.
|
9
|
- Pigeons were trained to categorize sound intensities, the 5 softest =
for
one key the 5 loudest for the other key.
- As the discrimination was acquired
- The curves for correct and incorrect responses separate.
- The asymptotes separate.
- The criterion is placed in a more optimal position.
|
10
|
- Eric Heinemann sought to account for the findings described by signal
detection theory by developing a model of discrimination learning and
categorization.
- Using computer simulation the
model provides a precise, quantitative description of the
observed behavior.
- We call the model the Natural Intelligence Model (NIM) in recognitio=
n of
insights the study of living organisms bring to our understanding of
intelligence--mechanisms that have evolved as organisms cope with the
issues of survival.
|
11
|
|
12
|
- During the first stage of learning, the presolution period, the pigeon discovers the
correlation between changes in stimulation and the outcome of its
behavior.
- During the second stage of learning, there is a gradual increase in
choice accuracy. We account for this by the gradual accumulation of relevant
information in exemplar memory.
|
13
|
|
14
|
- When a response (e.g. peck at an illuminated disk) is followed by a
motivationally significant event (a reinforcer) the sensation
experienced and the response made are placed at a random location in
exemplar memory—original estimates suggested 1200 such locatio=
ns.
- The record that previously occupied the location in which the new me=
mory
is placed is erased.
- This accounts for new
learning as well as forgetting.
|
15
|
- When a new stimulus is presented for identification a small sample of
records are drawn from exemplar memory.
- The retrieved records are not remembered exactly, e.g. a light inten=
sity
may be remembered as dimmer or brighter than that actually
experienced. This show=
s 2
records (green) for one choice and 3 (blue) for the second choice.
Notice that the means differ.
|
16
|
- In this illustration the decision is between one of two choices. The
probability density at the point of the current input is added for e=
ach
choice separately (e.g. green for the left key, blue for the right k=
ey)
.
- The response made is the one for which the probability density is
highest. This is higher (.42) for the distributions shown in blue (e.g. associated with a
right key choice ) tha=
n the
other key (.41). The right key is chosen.
|
17
|
- The degree to which the sensations produced by the stimuli
- overlap.
- Sample size – the number of items in working memory
- As numbers increase, errors decrease
- This figure shows the effects of sample size on proportion of cho=
ices
to one of two keys.
- Response confusion – errors due to confusion among choices
- Minimized by definin=
g the
required choice by easily discriminated key locations.
- Response confusion is a factor&nb=
sp;
when the decision requires discrimination among the response
manipulanda as well as the discriminative stimuli as in
“symbolic matching to sample.”
- See Blough (1996) for a comparison of left-right and symbolic
matching (Error Factors in Pigeon Discrimination and Delayed Matc=
hing
Journal of Experimental Psychology: Animal Behavior Proces=
ses)
|
18
|
|
19
|
|
20
|
|
21
|
|
22
|
- Outline drawings are represented&nb=
sp;
as dot-matrix patterns.
- The same algorithm used for categorizing diffuse stimuli is applied =
to
these patterns.
- Each point representing the stimulus to be identified is compared to
each point on records =
of
remembered stimuli. This process can be viewed in detail in EricR=
17;s
and my chapter in Cook’s (2001) cyberbook on Avian Visual
Cognition
|
23
|
- For humans the lines in the
- L-s=
hape
are easier to identify than the lines alone.
- The opposite is true for pigeons.
|
24
|
|
25
|
- NIM has enabled us to account for phenomena such as:.
- Relation between discriminability and the length of the PSP
- Probability and probabilistic discrimination learning
- Differences in accuracy as a function of discriminability of the
stimuli
- Stimulus identification, categorization and generalization
- Stimulus control by compounds of lights and sounds
- Choice among as many as nine alternatives.
- Blocking and overshadowing
- Identification and categorization of outline drawings
- Decreased accuracy in the presence of common elements
|