Efficient algorithms for universal denoising

The classical framework of context-tree models used in sequential decision problems such as compression and prediction has been recently generalized to a setting where the observations are multidimentional. Context set definitions tree representations and prunning algorithms have recently been extended from the classical unidirectional setting to a bidimensional one. In this case, it may be beneficial to consider contexts comprised of possibly different number of symbols from each direction. In this talk we present an efficient algorithm to implement this framework in a two dimensional setting, with special application to universal decoding.

This is a joint work with Marcelo Weinberger (HP Labs, California) and Fernando Fernandez (Universidad de la Republica, Uruguay).