
Random fluctuations in a measurement or computation result.Statistical noise refers to random fluctuations in a measurement or computation result. These fluctuations are inherent to any physical system and arise from the discrete nature of matter and energy, as well as from the finite precision of instruments. In science, noise is not just an annoyance; it can carry information about the underlying processes that generate it.
The study of noise began in the late 19th century with the observation of Brownian motion, explained by Einstein in 1905. The mathematical framework for noise was developed by Norbert Wiener and others in the mid-20th century, leading to the field of stochastic processes.
Noise arises from many sources: thermal agitation of electrons in a resistor (Johnson-Nyquist noise), the random arrival of photons at a detector (shot noise), or the stochastic nature of chemical reactions in a cell. In measurements, noise sets a fundamental limit on sensitivity, but statistical methods can extract signals from noisy data by averaging or correlating many repetitions. In computation, noise can be introduced by random number generators or by rounding errors, and it can affect the stability of algorithms.
Noise analysis is used in gravitational wave detectors to separate faint signals from background vibrations, in medical imaging to improve signal-to-noise ratios, and in machine learning to regularize models and prevent overfitting.
The hiss heard in a radio between stations is partly due to cosmic microwave background radiation, a faint noise left over from the Big Bang.


