Learning explicit categories
- Each instance (individual or episode) specifies values for lots of specific features.
- During learning,
- instances are clustered together on the basis of similarity in their feature values
- for each cluster of instances, an explicit category is formed by blending the
instances in the cluster:
- features whose values are shared (by a significant proportion of the instances)
are included in the category
- features whose values are not shared (by a significant proportion of the instances)
are excluded from the category
- Each resulting category is stored in long-term memory (LTM)
- At processing time,
- a test instance comes into the system
- the category or categories in LTM that best match the test instance are found
- missing values in the instance are filled in from the category/categories
Instance-based (case-based, exemplar-based) learning
- Each instance (individual or episode) specifies values for lots of specific features.
- During learning, each instance is simply stored in long-term memory (LTM)
- At processing time,
- a test instance comes into the system
- the instances in LTM that best match the test instance are found
- the matching instances are blended together into a category-like representation
- missing values in the test instance are filled in from the instance blend
