If we do not have a priori assumptions about the feature space, then all algorithms behave the same.
For example, suppose our computer has only two storage units. Storage is a special certificate in the feature space, and each storage unit is given a label. There are two kinds of labels. We give the label of one storage unit, and then predict the label of the other storage unit. Without assuming which of the two labels has a high probability, the probability of success is 50% regardless of which label is predicted by the algorithm
We believe that the samples with small feature gap are more likely to be the same class.
This work adoptsCC agreement, reprint must indicate the author and the link to this article