Address reduction blindly identifies non-random data series

We introduce a method of detecting data series (curves) which exhibit pattern without knowing what kind of pattern they contain. By partitioning the space of curves into neighbourhoods, we show that the curves with the shortest addresses are the most likely to result from simple underlying mechanisms. We show that address reduction is a bound on Kolmogorov complexity and is invariant over noise and one-to-one transformations. We use it to blindly identify gene expression profiles in yeast cell cycle and the segmentation clock, and to segregate humanand computer-generated random data.