OE-11.4 Classical & Fuzzy Sets

What this is and why it exists

Classical set membership is yes or no. Fuzzy set membership is a degree between zero and one, and that single change is how "the water is warm" becomes computable.

There is one distinction here that is muddled more often than any other in this subject, so it gets stated at the top.

Fuzziness is vagueness, not probability. A membership of 0.7 in "tall" does not mean a seventy percent chance the person is tall. It means the person is tall to a degree of 0.7 — the height is known exactly, and the *category* is what has soft edges.

The vocabulary

• Classical set — an element is in or out.
• Fuzzy set — an element belongs to a degree.
• Membership function — the rule assigning a degree to each value.
• Vagueness — the category has no sharp boundary.
• Probability — uncertainty about which of several definite outcomes holds.
• Support — the values with any membership above zero.
• Cardinality — the total membership summed over the set.
• Fuzzy relation — a degree of association between pairs of elements.

The mental model

Start with the failure that motivates all of it. Classical sets force a sharp boundary. Define "tall" as above one hundred and eighty centimetres and a person one millimetre shorter is entirely not tall. Nobody uses the word that way, and a controller built on such boundaries jumps abruptly as a measurement crosses one.

Fuzzy sets replace the boundary with a gradient. A membership function assigns each height a degree between zero and one. The category then has soft edges, and a system built on it responds smoothly.

Now the distinction, restated because it does the most work. Probability is about uncertainty over definite outcomes. A seventy percent chance of rain means it will either rain or not, and you do not yet know which. Fuzziness is about categories with soft edges. The height is measured exactly and known; the word "tall" is what admits of degree. When the rain arrives, the probability resolves to zero or one. The membership does not resolve, because there was never anything to resolve.

The operations follow from the classical ones by extension. Intersection takes the smaller of two memberships, union the larger, complement one minus the value. Most classical properties survive, and one famously does not: an element can belong to a set and its complement at once, both to some degree. That is not a defect, it is the point.

Relations extend the same idea to pairs — a degree of association rather than a yes or no connection. Cardinality becomes the total membership summed over the set, which is a real number rather than a count.

Membership functions come in standard shapes — triangular, trapezoidal, and smooth curves. The shapes are usually chosen for computational convenience and because a domain expert can describe them, not derived from data. Being clear that they are a modelling choice is part of using them honestly.

What you should now be able to explain or do

Contrast classical with fuzzy membership. State the difference between vagueness and probability, with an example of each resolving or not. Apply intersection, union and complement to fuzzy sets. Say which classical property fails and why that is intended. Compute the cardinality of a fuzzy set. Construct a membership function and say that its shape is a modelling choice.

Check yourself