discriminators

Discriminators refer to systems, models, or devices that distinguish between different categories or conditions. They appear in fields such as machine learning, statistics, and signal processing, and they can be physical circuits, software components, or mathematical functions.

In machine learning, a discriminator is a network or algorithm that attempts to tell real data apart from synthetic data generated by a counterpart model (the generator) within a generative adversarial network (GAN). It outputs a probability that an input is real and is trained to maximize correctness against the generator's outputs while the generator learns to fool it. Variants include patch-based discriminators and conditional discriminators. Stabilizing techniques include normalization and architectural choices.

In discriminant analysis, a discriminant function is used to classify observations into predefined groups. Linear Discriminant Analysis and Quadratic Discriminant Analysis construct decision boundaries based on feature distributions and prior probabilities. The method assumes labeled data and often multivariate normality; its results are probabilistic assignments and insight into feature contribution.

In signal processing, a discriminator (for example, a frequency discriminator or FM discriminator) extracts the instantaneous frequency or otherwise demodulates a modulated signal. It converts variations in frequency into a readable, baseband signal for further processing.

Outside technical usage, the word can denote a rule, criterion, or agent that differentiates between options or groups, sometimes raising concerns about bias or fairness when applied to real-world decisions.