Information Theoretical Approach to Attention in Geomagnetic Storm Forecasting: a Tutorial

Satellites and ground-based stations have recorded various types of data from the solar-terrestrial system during recent decades. The new type of particle detectors in SEVAN (Space Environmental Viewing and Analysis Network) project will be able to measure changing fluxes of most species of secondary cosmic rays, concurrently. Like the other sensor networks, SEVAN’s collected data includes a variety of indices from observed phenomena. Detecting especial events from this huge amount of assembled data, clarifies the essential need of developing methods to decrease the computation costs. In this tutorial, we try to provide a method to detect special events over a semantic framework, called attention control; the condition of readiness for such attention involves especially a selective narrowing or focusing of consciousness and receptivity. Generally, attention control is a way to escape from information bottlenecks, while selecting the most related parts of data to the target or the most critical events, due to the processing cost of each part. In first step, with no attention to the observation cost, selection of the most relative part of data to the especial events can extremely reduce the amount of data to be processed and relax the computational complexity problem. Our approach is based on Mutual Information (MI), as a generalization of correlation, form Information Theory which measures the amount of knowledge shared between two or more random variables. It is one of the most powerful and suitable criterion which can indicate any kind of relation, linear or nonlinear, between variables. Encountering a nonlinear system with complicated multi-dimensional dynamics requires MI, superior to correlation analysis, to recognize the existence of such nonlinear interrelations. It is also very efficient and meaningful to evaluate either relevance or redundancy of each variable when physical explanation of the selected variables is essential to increase the knowledge about the system.