Suppose we can afford taking a lot of samples, and for each sample, we fit the model and make a prediction f(x) about the signal f(x) at a given input x. What does the bias-variance tradeoff phenomenon tell us about the predicted value f (x)? The simpler the model, the closer the average value of f (x) (accross all samples) is to the true value f(x); however, the more likely will a single f (x) (i.e. for a single sample) be away from f(x). ○ The more complex the model, the closer the average value of f (x) (accross all samples) is to the true value f(x); however, the more likely will a single f (x) (i.e. for a single sample) be away from f(x).

Suppose we can afford taking a lot of samples, and for each sample, we fit the model and make a prediction f(x) about the signal f(x) at a given input x. What does the bias-variance tradeoff phenomenon tell us about the predicted value f (x)? The simpler the model, the closer the average value of f (x) (accross all samples) is to the true value f(x); however, the more likely will a single f (x) (i.e. for a single sample) be away from f(x). ○ The more complex the model, the closer the average value of f (x) (accross all samples) is to the true value f(x); however, the more likely will a single f (x) (i.e. for a single sample) be away from f(x).