Here is the smallest puzzle that trips up simple models. The Dataset has four rows, each with two inputs, x1 and x2, and a label that is either blue or amber. The pairs (0, 0) and (1, 1) are blue. The pairs (1, 0) and (0, 1) are amber.
On the plot, blue sits in two opposite corners and amber in the other two. A person sees the pattern at once: the label depends on whether the two inputs differ. The question is what kind of model can learn that.
💡Good to know
This pattern is known as XOR, short for exclusive or. It is a classic tiny test for models.