Variance is the error that comes from a model's sensitivity to small fluctuations in the training set: trained on a different sample, a high-variance model makes noticeably different predictions. High variance can cause an algorithm to model the random noise in the training data, which is overfitting.
Card 1 of 259. Answer: Variance is the error that comes from a model's sensitivity to small fluctuations in the training set: trained on a different sample, a high-variance model makes noticeably different predictions. High variance can cause an algorithm to model the random noise in the training data, which is overfitting.
About · Quizzes · History of AI · To suggest new questions or report an error send me a dm.
Variance is the part of a model's error that comes from its sensitivity to the particular training set: trained on a different sample from the same population, a high-variance model gives noticeably different predictions. Formally, at an input x, Var[f̂(x)] = E[(f̂(x) − E[f̂(x)])²], with the expectation over possible training sets.
Example: train the same algorithm on four different samples of 500 orders and ask each copy for the delivery time of the same order: 29, 31, 30 and 34 minutes. The mean is 31 and the variance ((−2)² + 0² + (−1)² + 3²)/4 = 3.5. If the true time is 25 minutes, the 6-minute gap between average prediction and truth is bias, not variance.
Unlike bias, variance can be estimated without knowing the true answer: retrain on bootstrap resamples and see how far predictions spread. Bagging turns that into a fix by averaging the copies. Keep variance apart from irreducible noise, the scatter of y around the true function: more data usually shrinks variance but never removes noise.
