Skip to main content
. 2021 Sep 28;10(9):giab055. doi: 10.1093/gigascience/giab055

Figure 5:

Figure 5:

Covariate shift: Inline graphic stays the same, but the feature space is sampled differently in the source and target datasets. A powerful learner may generalize well as Inline graphic is correctly captured [27]. Thus the polynomial fit of degree 4 performs well on the new dataset. However, an overconstrained learner such as the linear fit can benefit from reweighting training examples to give more importance to the most relevant region of the feature space.