I think it's interesting how with lossy compression it's less about optimizing for mathematical constructs (like min-norm solutions) and rather really about exploiting the way that we experience images.