Ugly duckling theorem

The ugly duckling theorem states that, without assumptions favoring particular properties, any two objects are equally similar because they share the same number of describable properties.

Connect

Feature space: A representation in which objects are described by values along a set of features. The theorem counts shared properties in a space whose features define how objects are described.
Satoshi Watanabe: A Japanese mathematician and information theorist who studied pattern recognition and inductive inference. Watanabe formulated the theorem and developed its implications for recognition.
Machine learning: Methods that enable computer systems to learn patterns from data. Learning systems must privilege some data structures or features to generalize successfully.
Minkowski distance: A family of distance measures defined over coordinate differences in a feature space. Choosing its coordinates and exponent imposes a particular geometry absent from neutral property counting.
Boolean algebra: An algebraic system built from logical operations such as AND, OR, and NOT. The proof treats property descriptions as combinations of predicates and their negations.
The Ugly Duckling: A Hans Christian Andersen fairy tale about a bird rejected for looking unlike its companions. Watanabe borrowed the tale’s title to illustrate how classification depends on chosen criteria.
Feature selection: The process of choosing which input variables a predictive model uses. Feature selection explicitly supplies the preferences absent from the theorem’s neutral setup.
Cosine similarity: A measure of similarity based on the angle between two vectors. It makes vector direction decisive rather than counting every describable property equally.
Predicate logic: A formal system for expressing claims about objects, their properties, and relations. Predicates provide the describable properties that the theorem counts.
Pattern recognition: The process of identifying regularities or categories in data and sensory input. The theorem challenges claims that pattern recognition can proceed without prior choices.