No abstract

The classical learning problem of the pattern recognition in a finite-dimensional linear space of real-valued features is studied under the conditions of a non-stationary universe. The training criterion of non-stationary pattern recognition is formulated as a generalization of the classical Support Vector Machine. The respective numerical algorithm has the computation complexity proportional to the length of the training time series.

Abstract.A data stream is an ordered sequence of training instances arriving at a rate that does not permit to permanently store them in memory and leads to the necessity of online learning methods when trying to predict some hidden target variable. In addition, concept drift often occurs, what means that the statistical properties of the target variable may change over time. In this paper, we present a framework of solving the online pattern recognition problem in data streams under concept drift. The framework is based on the application of the Bayesian approach to the probabilistic pattern recognition model in terms of logistic regression, hidden Markov model and dynamic programming.