No abstract

There are several tables with important data used in the calculus of different processes like machining tables, friction tables and thermodynamics processes tables, or as it is explored in this paper, the description of saturated water and steam table. We propose the generation of equations for describing the entire behavior of numerical values in a table using Genetic Programming (GP), when table data describes the variable behavior of a dependent function. This obtained equations simplify the calculus process without requiring several tables and allowing to work when tables are not available for a desired value of an independent variable, a common situation in thermodynamics. In this case it is tested the proposed algorithm for synthesizing the saturated water and steam table.

Recoloring it is a technique for changing the color an image resulting in a different new one. Recoloration is a common photo edition operation since digital images are around every media resource and several algorithms are used for editing these pictures, nevertheless, recent digital cameras have increased enormously the quantity of pixels for producing them. This increase in the size of digital images makes difficult the recoloring operation. In order to solve the recoloring problem, there had been applied several algorithms, some algorithms directly detect the color by performing transformations on color representations to different spaces where color is easily separated but this transformation require several no linear operations. On the other hand, numerical parameters on CNNs make than this approach cannot be trained or implemented on a mobile device, more over the time required for computing an input image will made that the processed pictures be delayed continually. Considering this limitation is proposed a specific short architecture for detecting a specific color in general objects using a feedforward neural network trained with gradient descent backpropagation with variable learning rate.

No abstract

No abstract

Tables are used in several areas where is required to show value responses, relationships, scores, percentages, and statistical results, among others. Tables increase accessing speed to information, but they need space for be presented and saving space could make that all the required values be not always included. Computer programs can replace tables, so that space not be required when presenting information, this has been explored since computers were developed presenting several algorithms. Fuzzy systems could be an alternative to generate programs for replace tables, more over this allow to get an expert system that knows the data in the table, but fuzzy systems require several numerical parameters to be tuned. In this paper is proposed the use of Batch Least Square Mamdani system for mimic tables in computer programs specifically applied in the AWG table which is a very common table used by engineers and electrical technicians.

No abstract

In this paper different competitive learning algorithms for self-organizing maps (SOM) are experimentally examined. The characterization of the results obtained is presented in terms of quality of SOM. The competitive learning algorithms evaluated through SOM are winner-takes-all, frequency sensitive competitive learning, and rival penalized competitive learning. Case study: their performance in the classification of partial discharges on power cables.