Introduction
Two subspaces tell the whole story of a linear map: the kernel — everything it crushes to zero — and the image — everything it can produce. Their dimensions always add up to the input dimension. This "conservation law" explains why information is lost when a model compresses data.