Memory-based Learning

• Lazy learning
• All encountered examples are stored in memory in a multi-dimensional array, positioned according to relevant Features
• New items are classified (comprehension) or generated (production) by searching for an example in memory that is closest to the target
• Because examplars are represented by their Features even novel forms can be classified
• A generalization of the knn (k-nearest neighbors) algorithm
• Don’t remove any infrequent or even solo forms. You might need the info
• Don’t trim down the number of examples of a frequent form you have in the model. This effects it.
• Learning is storing, classification is analogy
• multiple long-distance dependencies