Nowadays supercomputers allow us to solve numerical problems in 2D and 3D with high accuracy. But what to do with problems in higher dimensions which are typical in finance mathematics, stochastics, signal processing, data mining etc? Modern supercomputers are still very far from solving problems in e.g. 20 dimensional space. This is so-called curse of dimensionality. So the sparse data formats become important. Recent results in the sparse tensor approximation allow effective (fast and cheap) representation of some classes of high-dimensional operators with the needed storage requiremen O(dnlog n) instead of O(n^d), where d is the dimension.