Nonuniform Wavelet Approximation in Databases
In this project, we introduce new Haar wavelet synopses which approximate the time-varying data in databases with nonuniform accuracy, which is specified by the biased queries from users (weights). Our data synopses are designed to be efficient not only in generation but also in maintainance when dealing with time varying data. The generation cost is linear in term of both time and space, and the update cost is sublinear. The accuracies of our synopses are validated against other linear methods by using both synthetic and real data sets.
This work is described in the paper "Nonuniform Compression in Databases with Haar Wavelet"