Recurrent Backpropagation Networks Research Articles

Evaluation of protein surface roughness index using its heat denatured aggregates

AbstractRecent research works on potential of different protein surface describing parameters to predict protein surface properties gained significance for its possible implication in extracting clues on protein's functional site. In this direction, Surface Roughness Index, a surface topological parameter, showed its potential to predict SCOP-family of protein. The present work stands on the foundation of these works where a semi-empirical method for evaluation of Surface Roughness Index directly from its heat denatured protein aggregates (HDPA) was designed and demonstrated successfully. The steps followed consist, the extraction of a feature, Intensity Level Multifractal Dimension (ILMFD) from the microscopic images of HDPA, followed by the mapping of ILMFD into Surface Roughness Index (SRI) through recurrent backpropagation network (RBPN). Finally SRI for a particular protein was predicted by clustering of decisions obtained through feeding of multiple data into RBPN, to obtain general tendency of decision, as well as to discard the noisy dataset. The cluster centre of the largest cluster was found to be the best match for mapping of Surface Roughness Index of each protein in our study. The semi-empirical approach adopted in this paper, shows a way to evaluate protein's surface property without depending on its already evaluated structure.

Predicting the exchange traded fund DIA with a combination of genetic algorithms and neural networks

Abstract We evaluate the performance of a heterogeneous mixture of neural network algorithms for predicting the exchange-traded fund DIA. A genetic algorithm is utilized to find the best mixture of neural networks, the topology of individual networks in the ensemble, and to determine the features set. The genetic algorithm also determines the window size of the input time-series supplied to the individual classifiers in the mixture of experts. The mixtures of neural network experts consist of recurrent back-propagation networks, and radial basis function networks. The application of genetic algorithm on the heterogeneous mixture of powerful neural network architectures shows promise for prediction of stock market time series. These highly non-linear, stochastic and highly non-stationary time series have been found to be notoriously difficult to predict using conventional linear statistical methods. In this paper, we propose a biologically inspired methodology to tackle such hard problem using a multi-faceted solution.

Connectionist simulation of quantification skills

The study of numerical abilities, and how they are acquired, is being used to explore the continuity between ontogenesis and environmental learning. One technique that proves useful in this exploration is the artificial simulation of numerical abilities with neural networks, using different learning paradigms to explore development. A neural network simulation of subitization, sometimes referred to as visual enumeration, and of counting, a recurrent operation, has been developed using the so-called multi-net architecture. Our numerical ability simulations use two or more neural networks combining supervised and unsupervised learning techniques to model subitization and counting. Subitization has been simulated using networks employing unsupervised self-organizing learning, the results of which agree with infant subitization experiments and are comparable with supervised neural network simulations of subitization reported in the literature. Counting has been simulated using a multi-net system of supervised static and recurrent backpropagation networks that learn their individual tasks within an unsupervised, competitive framework. The developmental profile of the counting simulation shows similarities to that of children learning to count and demonstrates how neural networks can learn how to be combined together in a process modelling development.

Experimental studies of network parameters and operational variables on recurrent backpropagation neural network adaptive control of penicillin acylase fermentation by Arthrobacter viscosus

This work is to investigate the on-line control of the fermentation by Arthrobacter viscosus. This species of bacteria can secrete penicillin acylase which is a key enzyme in pharmaceutical industry. The growth of more cells during the fermentation will obtain more enzyme. Both the enzyme activity and the cell growth are rather sensitive to the change of pH. Once the pH during a fermentation is not properly controlled, the decay of cells' activity will irreversibly occur. Two peristaltic pumps for supplying acidic and basic solutions, respectively, were connected for the regulation of pH. With superior ability in identification and prediction of dynamic time series, recurrent backpropagation network (RBPN), instead of conventional controllers, was used as the adaptive controller model for the fermentation with dynamic characteristics. Based on a 1-3-1 BPN, a corresponding 4-4-1 RBPN was determined. The deviation of the pH measured at current time from the set point of 7.0, denoted as Δ pH(t), was chosen as the input node of the network controller. The output node of this network controller was the predicted flow rate of the peristaltic pump for next control time interval. Such a model was operated by two phases. During the first phase, the network was set as the process model and trained by a fixed set of on-line acquired data. During the second phase, the network was stopped learning and switched to become a predictor, the predicted control action was hence obtained. The optimum sampling time was determined experimentally. To enhance the effective computation of this network, the number of training data was limited. A moving-window type of supplying training data to the network was applied for the on-line learning. The window size was also determined for each learning. With properly chosen network parameters as well as operation conditions, pH of the fermentation was thus well controlled by the RBPN controller.

Radar image segmentation using recurrent artificial neural networks

This paper discusses the application of artificial neural networks to the segmentation of Doppler radar images, in particular the detection of oil spills within sea environments, based on a classification of radar backscatter signals. Best results have been achieved with recurrent backpropagation networks of an architecture similar to that of Elman's Simple Recurrent Network (1990). These recurrent networks are shown to be very robust to variations in both sea state (weather conditions) as well as illumination distance, and their performance is analysed in further detail.