great detail is needed for planning

1
Review more detail vs. less detail

• great detail is needed for planning executing
  current responses
• some detail about context of new learning can be
  emcoded with that learning, and be a cue.
• less detailed representation is more useful for
  learning general, widely-applicable lessons for
  future reference

2
Categorization
My dog sleeping. My dog. All golden retrievers.
All dogs. All canines. All mammals Each of
these is a category. Categorization is the
process of deciding which details matter, and
which dont, for some purpose.
3
Advantage we gain by categorizing things

• Bruner, Goodnow, Austin (1956)
• reduce complexity of environment
• generalize lessons
• guide choice of response
• make hierarchical knowledge available

4
Two questions about categories

• 1. What is the structure of natural categories
  like?
• That is a question about the world.
• How is information about natural categories
  represented in memory?
• That is a question about your mind.

5
1. The structure of natural categories

• Is this a question about the world, or about us?
• Which two are most similar sheep, goats, cows?
• To some extent, structure of natural categories
  is given by the world. To some extent, it is
  impressed upon the world by human cognition.

6
1. The structure of natural categories

• Most important work was done by Eleanor Rosch.
• Hierarchy.
• Rosch argued that our categorical knowledge is
  organized in a hierarchy superordinate, basic
  level, subordinate.
• This is a relation of containing.

7
Mammal Dogs Cats Horses Collie
Airedale Persian Siamese Arabian
There are three levels in this hierarchy. They
are not all equal.
8
Basic level

• The basic level is the most important one
• Between superordinate and subordinate
• Its the one we use when we name an object.
• Its the one children learn first.

9
Shared characteristics
Superordinate items in a category can be very
different from each other. E.g., table, chair,
lamp. Subordinate items in different
categories can be very similar to each other.
E.g., dining room chair, patio chair. Basic
level items similar to others in category,
different from those in other categories.
10
Review - Hierarchy

• Things in the world present themselves in a
  hierarchy of levels of categorization
• At basic level, items in a category look like
  each other but not like members of other
  categories.
• Basic level is one used spontaneously in naming
  objects.

11
Roschs second contribution - Typicality

• Rosch argued that some members of a category are
  better than others that is, more typical.
• such members have family resemblance.
• typical members are similar to other members,
  unlike non-members of category

12
2. The representation of natural categories

• Some categories are natural e.g., mammals.
• Some categories are artificial e.g., all
  animals that weigh more than 100 lbs.
• Some are functional things to bring out of the
  house in the event of fire.

13
2. The representation of natural categories
Four models Prototype Feature frequency Nearest
Neighbour Average distance
14
Prototype models

• A prototype is a typical member of a category
• Prototype theories say that, through experience,
  we create a central example of each category.
• Prototype may exist only in your mind (e.g., not
  as an actual object in the world).

15
Feature frequency

• Categorization is based on how many features the
  to-be-classified object shares with each of the
  available categories.
• E.g., a whale shares breathes air and gives
  birth to live young with mammals. It shares
  lives in the ocean and moves by tail and
  flipper action with fish.
• So we should expect confusion about whales

16
Nearest Neighbour
New object is compared with each exemplar of each
stored category. Compute difference between
object and each exemplar in each category. New
object is classified in category containing
object it is most similar to (smallest
difference).
17
Average Distance
Same comparison of new object to all stored
exemplars of categories as in nearest
neighbour. Decision based on which category has
smallest average distance from the new
object. Compare with Nearest Neighbour model
here, it is average distance for the category,
not just which exemplar is closest, that counts.
18
Review models of representation

• Two fundamentally different views of how we store
  category information
• Prototype model
• what is generally true about something is
  stored and available when needed
• this view emphasizes abstract representations

19
Review models of representation

• N.N. and A.D. models
• what is generally true is not stored, but
  computed when needed.
• these views emphasize storage of individual
  experiences with objects