Structural and parametric data representation using the second order optimization method

Abstract

Working with various data sources in real-time requires approaches capable of adaptive parameters tuning. We propose algorithms that represent dynamic data streams in apriori defined structures. The algorithms are based on the certain error minimization. The used method is Newton's method, which is appropriate because of its high convergence. At every step, when the new data are received we make corrections to the unknown parameters vector by solving differential equations systems. Initial values are selected using estimates obtained from the practical stability theory. The computational experiment was conducted to compare models based on the first and second order optimization approaches. It confirms the effectiveness of our approach.