Efficient Algorithms for Constructing Optimal Bi-directional Context Sets

Bi-directional context sets extend the classical context-tree modeling framework to situations in which the observations consist of two tracks or directions. In this paper, we study the problem of efficiently finding an optimal bi-directional context set for a given data sequence and loss function. This problem has applications in data compression, prediction, and denoising. The main tool in our construction is a new data structure, the compact bi-directional context graph, which generalizes compact suffix trees to two directions.