multi layer perceptron

Many researchers proved that hybrid models have better results in comparison with independent models. A combination of different methods could enhance the accuracy of time series prediction. Hence, this research used the hybrid of three methods of chaos theory, multi-layer perceptron and metaheuristic algorithm to increase the power of the model forecasting. Artificial neural networks have properly considered complex nonlinear relations and are good comprehensive approximators. Multi-objective evolutionary algorithms such as multi-objective particle swarm optimization are good at solving multi-objective optimization issues. This algorithm organized the combination of parent and children populations by elitist strategy, decreased the messy comparing factors to improve the solution variety and avoided to use of niche factors. Chaos theory controls the complexities of stochastic systems. So, this research offers Tehran Stock Exchange Index (TSEI) prediction by a hybrid model of chaos theory, multi-layer perceptron and metaheuristic algorithm. The results show that in perceptron-based mode, RMSE measures are gradually increased in all intervals. The continuous decrease of RMSE shows that the perceptron-based model could show consistency with the whole data flow. This matter could offer a better learning and consistency process by perceptron-based models to predict stock prices, as this type of learning could apply more experiences for forecasting future behaviour in order to change the system content.

This paper aims to present a useful method for magnifying images, for which it is necessary to group the pixels and define the borders. In the proposed method, images are first partitioned using suitable segmentation algorithms and then artificial neural networks (ANNs) are applied to magnify each segment individually. In the ANNs applied, training is performed using, as input, a down-sampled form of the same image to be magnified. This type of training results in a high quality zoom in each segment since the pixels in an individual segment have very close features. Evaluation results on several images verifies the higher efficiency of the proposed method than other recently developed image zooming methods.

In this paper, we introduce a Takagi-Sugeno (TS) fuzzy model which is derived from a typical Multi-Layer Perceptron Neural Network (MLP NN). At first, it is shown that the considered MLP NN can be interpreted as a variety of TS fuzzy model. It is discussed that the utilized Membership Function (MF) in such TS fuzzy model, despite its flexible structure, has some major restrictions. After modifying the MF, we introduce a TS fuzzy model whose MFs are tunable near and far from focal points, separately. To identify such TS fuzzy model, an incremental learning algorithm, based on an efficient space partitioning technique, is proposed. Through an illustrative example, the methodology of the learning algorithm is explained. Next, through two case studies: approximation of a nonlinear function for a sun sensor and identification of a pH neutralization process, the superiority of the introduced TS fuzzy model in comparison to some other TS fuzzy models and MLP NN is shown.