1. Discrimination learning: Introduction
Discrimination learning in nature
Simple discriminations
How are they learned?
3 theories
Discriminating complex stimuli in the lab
Do animals learn concepts?

2. Discriminating prey from background
Some things that get eaten are camouflaged
Predators’ ability to learn difficult discriminations  evolution of camouflage

3. This moth matches the kind of tree bark it rests on.
On the web: link to experiments in which bluejays look for moths like this.

4. Discriminating good vs. poisonous food
Monarch
butterflies
Bright colors
help predators
learn not to
attack them.

5. Avoidance of monarch generalizes to harmless butterflies that look similar
Predators’ discrimination and generalization drives the evolution of mimicry.

6. More mimicry: flies that look like bees
In each row, a bee
or wasp is on the left.
The rest are harmless
flies that evolved to look like bees
and wasps.

7. How are discriminations learned?
Simple discrimination training methods
Three theories
Spence
Rescorla & Wagner
Pearce

8. Example of a simple discrimination
 reinforcement (S+)
 no reinforcement (S-)

9. What’s going on? Spence’s theory
S+ S-
Excitatory associative strength generalizes S+
Associative strength
Stimulus similarity

10. Spence: Inhibition generalizes less than excitation
Stimulus similarity S+ S-
Associative strength

11. Spence: Responding determined by net strength
net = difference between the 2 gradients
Stimulus similarity S+ S-
Associative strength

12. Is discrimination absolute or relational?-
Train
What was learned?
absolute values (“dark gray is positive”)
relationship (“the darker one is positive”)
Transposition test vs. +
Absolute: choose the old S+
Relative: choose the darker, not the old S+

13. Spence can predict transposition
Train
Test
Test stimulus has more positive strength than old S+ - +
Stimulus similarity S+ S-
test

14. Difficulty with Spence’s model
Idea of generalization is very powerful
But Spence assumed different stimuli present at the same time were independent
Different from Rescorla and Wagner
Spence gets into trouble with compound stimuli….

15. Feature-positive discrimination
A+B  US, B  no US
 food nothing
A (the light) predicts food; tone does not
ALL positive strength should go to the light
NO CR to B (the tone)
But Spence’s model predicts a CR to B
Rescorla & Wagner is the better model here

16. A possible problem for Rescorla-Wagner
A US, B  US A+B  no US
 food  food  nothing
This should be impossible for animals to learn!

17. A+, B+, AB- can be learned: How?
% trials with a CR
trials

18. Maybe AB has a unique configural element i.e. AB really = ABC
(whole is more than the sum of parts)
“C” will be a strong inhibitor here
A+, B+
% trials with a CR
AB (C)-
trials

19. Pearce causes trouble for R&W: a more complex kind of discrimination
AB +
ABC -
Two ways to look at these stimuli:
Sums of elements (R-W)
Unique configurations (Pearce)

20. Pearce’s configural theory
A AB+ ABC-
Animals encode compounds as wholes.
Generalization occurs between a whole and its parts because they are similar
Here, ABC is more like AB than like A
ABC vs. AB should be learned slower than A vs. ABC

21. Pearce’s configural theory
A AB+ ABC-
R-W predicts that C (red) will become inhibitory
A(yellow) and B (black) will be excitors
AB will be more excitatory than A alone

22. Results:Support Pearce
A+, AB+, ABC-

23. Three Theories of Simple Discrimination Learning: Summary
Spence
S+ and S- generalization gradients
transposition
Configural discriminations
Rescorla & Wagner: analyze as elements
Pearce: configural theory
Each correct for some situations (book)

24. More complex stimuli: Category discriminations
S+ any slide with a tree
S- any slide with no tree
Train on 40 S+, 40S-
Test with novel slides
Birds perform better than chance on novel slides
i.e. they transfer
What is going on?

25. Category discrimination: What is learned?
Category discrimination (or classification) = different behavior toward things in different categories.
This is what the birds are doing.
Concept = cognitive structure that may underlie category discrimination.
Have the birds learned a concept?
There are other possibilities…...

26. Memorizing slides plays a role
Pigeons learn small irrelevant details
They can memorize 320 individual slides
Probably don’t see the slides as pictures
Memorizing and generalizing both contribute to acquisition and transfer
Data on next slide

27. Person, flower, car, or chair? Old and new slides mixed
Birds perform above chance on new slides (generalizing)
Better on old ones (memorizing)
% correct : old = black 

28. What is learned in category learning? Simple associations
Feature theory: Excitation and inhibition to features (e.g. green patches, leafy textures)
Transfer because features are shared
Exemplar theory: Excitation and inhibition to whole exemplars as configurations
Transfer reflects stimulus generalization

29. What is learned in category learning? More than simple associations
A prototype
Average or typical category member
Transfer reflects similarity to the prototype
A concept
Functional categories and mediated generalization (in the book)

30. “Animals”:
Training
exemplars
Transfer:
based
on common
features
Feature learning
4 legs
Eyes
Ears
Tail
Fur
Etc.

31. “Animals”:
Training
exemplars
Transfer: based on
similarity to the exemplars
Exemplar theory
all exemplars (configurations) memorized

32. “Animals”:
Training
exemplars
Transfer: compare to the prototype
Prototype learning
Representation of
a kind of average animal

33. What is learned in category discrimination? Conclusions
No good evidence for processes other than association formation
Associations could involve features or whole exemplars
Probably not prototypes
More in the book