high-dim features require exponentially more data to reliably estimate disances solution: dim red, i.e. singular value decomposition (SVD)