Swarm intelligence (SI) algorithms are handy tools for solving complex optimization problems. When problems grow in size and complexity, an increase in population or number of iterations might be... Show More

Swarm intelligence (SI) algorithms are handy tools for solving complex optimization problems. When problems grow in size and complexity, an increase in population or number of iterations might be required in order to achieve a good solution. These adjustments also impact the execution time. This article investigates the trade-off involving population size, number of iterations and problem complexity, aiming to improve the efficiency of SI algorithms. Results based on a parallel implementation of Fish School Search show that increasing the population size is beneficial for finding good solutions. However, we observed an asymptotic behavior, i.e. increasing the population over a certain threshold only leads to slight improvements. Furthermore, the execution time was analyzed.

In this article, the authors propose a new version of Fish School Search Algorithm named FSS-CS. This release has three significant changes. First, it has an improved feeding mechanism to enhance... Show More

In this article, the authors propose a new version of Fish School Search Algorithm named FSS-CS. This release has three significant changes. First, it has an improved feeding mechanism to enhance the barycenter calculation. Secondly, it promotes exploration by using a state-of-art, non-greedy strategy. Finally, it incorporates a promising existent elliptic step decay. The authors assessed the proposal in ten benchmark optimization problems to evaluate the performance. The results show that the proposed version outperformed in most cases the FSS versions for mono-modal optimization.

The excessive exposure to certain kinds of acoustic noise can lead to health problems. To avoid this situation, the use of noise attenuation devices is a standard solution. Among those devices, the... Show More

The excessive exposure to certain kinds of acoustic noise can lead to health problems. To avoid this situation, the use of noise attenuation devices is a standard solution. Among those devices, the active noise control (ANC) systems have gained prominence over the years, mainly due to the technological development and costs reduction of electronic components. Despite good performance of ANC concerning low-frequency noise attenuation, the convergence speed for this kind of system is still an important issue when it deals with real-time applications in dynamic environments. This article presents an alternative solution to accelerate the active attenuation system response. This solution is based on the use of sets of coefficients, which are employed during the adaptive filter initialization and are obtained via a training process with particle swarm optimization (PSO). Two objective functions were tested: one based on the response time itself and the other one based on the magnitude reduction of the residual noise. The coefficients obtained through this process provided response time reductions up to 98.3% concerning adaptive filters initialized with null coefficients. The article is an extended version of the conference paper Accelerating the Convergence of Adaptive Filters for Active Noise Control Using Particle Swarm Optimization, published in LA-CCI 2017.

The accumulation of secretions in the airways of ventilator-dependent patients is a common problem, and if not detected and treated in due time, it greatly increases the risk of infections and... Show More

The accumulation of secretions in the airways of ventilator-dependent patients is a common problem, and if not detected and treated in due time, it greatly increases the risk of infections and asynchrony. Unfortunately, cardiogenic oscillation modifies the flow signal shape that can confuse clinical staff and modern lung ventilators. In this article, the authors use an artificial immune system algorithm in a pre-processed flow signal. The authors' approach was able to automatically detect the presence or absence of airway secretions, even if the sample contains the influence of cardiogenic oscillation. The training and validation of the algorithm was carried out using a database containing flow signals of 457 respiratory cycles, obtained from three patients in different ventilation modes. The algorithm trained with 60% of the base cycles, was able to achieve specificity and sensitivity above 0.96 in the classification of the remaining cycles of the base.