Contributions to the Study of Dynamics in Information Theory: Applications in Dynamic Clustering
Information Theory, Dynamic Clustering, Dynamic Processes.
Information Theory is a branch of mathematics, more specifically probability theory, that studies information quantification. Recently, several researches have been successful with the use of Information Theoretic Learning (ITL) as a new technique of unsupervised learning. In these works, information measures are used as criterion of optimality in learning. In this work, we will analyze a still unexplored aspect of these information measures, their dynamic behavior. The main objective of this work is to investigate the use of measures of information theory in the context of dynamic processes. For this, the same was done in 3 (three) distinct phases. In the first phase we investigated the presence of dynamics in the information in the processes. As a source of dynamic information, videos with different characteristics were used. The second phase presents a new representation for dynamical processes by state space called Information State Representation. In this representation, the states of the system are described as a function of the information measures of the system. To validate this new form of representation, some experiments were carried out with videos aiming at evaluating its quality when submitted to different dynamic aspects. In the third phase, we investigated the use of measures based on information theory within the area of dynamic clustering. The objective in this phase was to compare the performance of the use of measures of information theory with traditional measurements in the operations of merge and split between clusters. The results obtained in all the phases were quite satisfactory meeting the objectives proposed in the work.