In this paper we studied the research activity of Brazilian Institutions for all sciences and also their performance in the area of physics between 1945 and December 2008. All the data come from the Web of Science database for this period. The analysis of the experimental data shows that, within a nonextensive thermostatistical formalism, the Tsallis q-exponential distribution N(c) can constitute a new characterization of the research impact for Brazilian Institutions. The data examined in the present survey can successfully be fitted by applying a universal curve namely, <img border=0 width=32 height=32 src="../../../../../../../img/revistas/bjp/v39n2a/a31tex01.gif" align=absmiddle>with q <img border=0 width=32 height=32 src="../../../../../../../img/revistas/bjp/v39n2a/a04tex01.gif" align=absmiddle>4/3 for all the available citations c, T being an "effective temperature". The present analysis ultimately suggests that via the "effective temperature" T, we can provide a new performance metric for the impact level of the research activity in Brazil, taking into account the number of the publications and their citations. This new performance metric takes the "quantity" (number of publications) and the "quality" (number of citations) for different Brazilian Institutions into account. In addition we analyzed the research performance in Brazil to show how the scientific research activity changes with time, for instance between 1945 to 1985, then during the period 1986-1990, 1991-1995, and so on until the present. Finally, this work intends to show a new methodology that can be used to analyze and compare institutions within a given country.

In this paper we apply an evolving stochastic method to construct simple and effective Artificial Neural Networks, based on the theory of Tsallis statistical mechanics. Our aim is to establish an automatic process for building a smaller network with high classification performance. We aim to assess the utility of the method based on statistical mechanics for the estimation of transparent coating material on security papers and cholesterol levels in blood samples. Our experimental study verifies that there are indeed improvements in the overall performance in terms of classification success and at the size of network compared to other efficient backpropagation learning methods.