Generalization

Generalization is a process in which confusing details of knowledge are ignored to provide for a clearer big picture. Generalization assists comprehension and reasoning. It is vital for human intelligence. It is an inherent property of neural networks and shows up even in primitive animals.

For example, if we make a complex and detailed mental picture of a dog. This picture can quickly lose its detail, and soon we will remember an encounter with a specific breed, e.g. a golden retriever. Generalization makes it easy to convert data-rich knowledge into data-sparse knowledge of higher applicability.

Forgetting plays a vital role in the process of generalization. We keep learning facts, and we keep forgetting facts. This endless, and seemingly futile process contributes to forming models via generalization. This is how learning of seemingly trivial facts may contribute to intelligence.

Generalization underlies the crystallization of the concept network of the brain: conceptualization. In a concept network, generalization occurs as an intersection of activations of a concept map in different contexts. Repeated activation of concept links leads to stabilization. Less often used links are subject to forgetting.

For more see: Abstract knowledge

This glossary entry is used to explain texts in SuperMemo Guru series on memory, learning, creativity, and problem solving

Abstract knowledge terminology

• generalization: a process in which detail is ignored to reveal a deeper structure. The term overlaps with abstraction, conceptualization, inductive reasoning, modeling, theorization, categorization, conclusion, unification, colligation, de-concretization, pattern extraction, pattern separation, and more. Example: Trump is a winner is a gross generalization that ignores Donald Trump's failures
• concept: a generalization of a set of objects/nouns. It overlaps with idea, entity, notion, group, etc. For example, animal is a concept derived from objects such as specific cats, birds, etc. Perhaps this should also include: property, attribute, quality, etc. i.e. the abstraction of object characteristics (e.g. the concept of yellowness)
• rule: a generalization of an observed regularity. It overlaps with formula, theorem, principle, proposition, law, statement, and more. Example: "no pleasure, no good learning" is a fundamental law of learning. It is an example of a general rule that determines learning strategies
• model: set of rules that apply to a specific phenomenon. It overlaps with theory, metaphor, opinion, schema, view, (concept) map, (formal) system, and more. Example: jigsaw puzzle metaphor of learning is a model of how knowledge coherence emerges in the process of learning
• abstractness: universality of a concept or a rule, e.g. 2 apples and 2 apples add up to 4 apples is less abstract (i.e. more concrete or more specific) than 2+2=4
• applicability: usefulness of a rule or model. It overlaps with usability. Example: 2+2=4 is useful in counting apples, but not-too-helpful in memorizing song lyrics
• abstract knowledge: well-generalized, highly applicable knowledge that is conceptual/abstract in nature. It overlaps with: "big picture", set of rules/formulas, abstract set, theory, etc. Example: mathematics is the queen of abstract knowledge
• abstract thinking: conceptual computation on rich and complex abstract knowledge