On context-tree prediction of individual sequences

TL;DR

This paper investigates the use of context-tree methods for universal prediction of individual sequences, analyzing how the growth rate of contexts affects prediction performance and proposing an optimal algorithm for sublinear growth rates.

Contribution

It introduces a universal context-tree prediction algorithm that performs optimally when the number of contexts grows sublinearly with sequence length, and establishes the linear growth rate as a critical threshold.

Findings

The critical growth rate of contexts is linear in sequence length.

The proposed algorithm achieves near-optimal performance for sublinear growth rates.

Linear growth in contexts prevents universal prediction performance.

Abstract

Motivated by the evident success of context-tree based methods in lossless data compression, we explore, in this paper, methods of the same spirit in universal prediction of individual sequences. By context-tree prediction, we refer to a family of prediction schemes, where at each time instant $t$, after having observed all outcomes of the data sequence $x_1,...,x_{t-1}$, but not yet $x_t$, the prediction is based on a ``context'' (or a state) that consists of the $k$ most recent past outcomes $x_{t-k},...,x_{t-1}$, where the choice of $k$ may depend on the contents of a possibly longer, though limited, portion of the observed past, $x_{t-k_{\max}},...x_{t-1}$. This is different from the study reported in [1], where general finite-state predictors as well as ``Markov'' (finite-memory) predictors of fixed order, were studied in the regime of individual sequences. Another important difference between this study and [1] is the asymptotic regime. While in [1], the resources of the predictor (i.e., the number of states or the memory size) were kept fixed regardless of the length $N$ of the data sequence, here we investigate situations where the number of contexts or states is allowed to grow concurrently with $N$. We are primarily interested in the following fundamental question: What is the critical growth rate of the number of contexts, below which the performance of the best context-tree predictor is still universally achievable, but above which it is not? We show that this critical growth rate is linear in $N$. In particular, we propose a universal context-tree algorithm that essentially achieves optimum performance as long as the growth rate is sublinear, and show that, on the other hand, this is impossible in the linear case.