Discriminator

A discriminator is any mechanism, device, or algorithm that distinguishes between two or more categories, signals, or values. The term emphasizes the act of discrimination or classification, rather than generation or synthesis, and the specific implementation varies across disciplines.

In machine learning and artificial intelligence, a discriminator is a component of a generative adversarial network (GAN). It is trained to differentiate real data from data produced by the corresponding generator. The two networks compete in a zero-sum game, with improvements in the discriminator driving advances in the generator and vice versa.

In signal processing, a discriminator refers to a demodulator circuit or algorithm that extracts information from a carrier signal by evaluating a property such as frequency or phase. Examples include frequency discriminators and phase discriminators used in demodulation of frequency- or phase-modulated signals, like FM radio systems or communication channels.

In statistics and pattern recognition, the term can describe a discriminant function or classifier that assigns observations to predefined categories. Methods such as linear discriminant analysis or more complex machine-learning classifiers function as discriminators by determining decision boundaries that separate classes based on observed features.

Beyond these domains, the concept appears in various engineering and computational contexts where a system must decide between alternatives, filter inputs, or route information accordingly. The specifics of a discriminator—its inputs, outputs, and performance criteria—depend on the application, but the common goal is to separate or identify distinct groups or signals reliably.