Concepts and Categories

  1. What is a rule-based model?
    • Evaluate rule (i.e. A if square, else B) - categories are defined by logical rules at every trial.
    • First try simple rule, but if rule rejected (bc of counterexamples), other rules are tested until get the right rule.
  2. What are rules?
    Concise representation of a category, but s/t more elaborate would be better/more descriptive.
  3. Disadvantages to rule-based theory?
    • Potentially useful info is discarded.
    • Most everyday categories don't seem to be describable by a tractable (manageable/controllable) rule.
    • If all category = a rule determining membership, then all members will be equal (not true - robin is more typical bird than penguin).
  4. Example of rule-based theory
    "A radiologist using rule-based categorization would observe whether specific properties of an X-ray image meet certain criteria; for example, is there an extreme difference in brightness in a suspicious region relative to other regions? A decision is then based on this property alone."
  5. What is prototype theory?
    • Mode of categorization, where some members of a category are more central than others.
    • E.G. when asked to give an example of the concept furniture, chair is more frequently cited than, say, stool.
  6. How are prototype based models formed?
    • Compare similarity to class centroids (i.e. A closest to A prototype, else B)
    • Encode central tendency (mean) of each category, comparing distance b/w 2 prototypes.
  7. What is the prototype of a category?
    • Summary of all its members.
    • Stimulus closest to group average on all dimensions is used as representative.
    • This is updated as new words are encountered - prototype shifts to account for it.
    • Helpful when no dimension is the key defining one item from category.
  8. If 2 items are close together in a plot, what does this mean?
    • More similar
    • Most typical items are closest to prototype
  9. Advantages of prototype based models
    • Useful when no individual stimulus dimension unambiguously distinguishes the categories
    • Efficient - store one vector per category
    • Accounts for typicality, including previously unseen exemplars being classified as most typical of the category
  10. Disadvantages to prototype based model?
    • Doesn't retain enough info about examples encountered in learning (e.g. frequency)
    • Misses anything not coded by mean: e.g. variance and frequency - if one category appears 10x more often than another you should assign ambiguous stimuli to the more frequent category (i.e. have to guess frequency based on base rate of stimulus working with)
    • In general, prototype models can only be used to learn category structures that = linearly separable (when a line can be drawn that separates all members of two categories).
Card Set
Concepts and Categories