Yes / A 9 x 18 x 1 feed-forward neural network (NN) model trained using a resilient back-propagation algorithm and early stopping technique is constructed to predict the shear strength of deep reinforced concrete beams. The input layer covering geometrical and material properties of deep beams has nine neurons, and the corresponding output is the shear strength. Training, validation and testing of the developed neural network have been achieved using a comprehensive database compiled from 362 simple and 71 continuous deep beam specimens. The shear strength predictions of deep beams obtained from the developed NN are in better agreement with test results than those determined from strut-and-tie models. The mean and standard deviation of the ratio between predicted capacities using the NN and measured shear capacities are 1.028 and 0.154, respectively, for simple deep beams, and 1.0 and 0.122, respectively, for continuous deep beams. In addition, the trends ascertained from parametric study using the developed NN have a consistent agreement with those observed in other experimental and analytical investigations.

yes / A 9 × 18 × 1 feed-forward neural network (NN) model trained using a resilient back-propagation algorithm and early stopping technique is constructed to predict the shear strength of deep reinforced concrete beams. The input layer covering geometrical and material properties of deep beams has nine neurons, and the corresponding output is the shear strength. Training, validation and testing of the developed neural network have been achieved using a comprehensive database compiled from 362 simple and 71 continuous deep beam specimens. The shear strength predictions of deep beams obtained from the developed NN are in better agreement with test results than those determined from strut-and-tie models. The mean and standard deviation of the ratio between predicted capacities using the NN and measured shear capacities are 1·028 and 0·154, respectively, for simple deep beams, and 1·0 and 0·122, respectively, for continuous deep beams. In addition, the trends ascertained from parametric study using the developed NN have a consistent agreement with those observed in other experimental and analytical investigations.

There are multiple factors that influence auditory steaming. Some, like frequency separation or rate of presentation, have effects that are well understood while others remain contentious. Human behavioural studies and event-related potential (ERP) studies have shown dissociation between a pre-attentive sound segregation process and an attention-dependent process in forming perceptual objects and streams. This thesis first presents a model that synthetises the processes involved in auditory object creation. It includes sensory feature extraction based on research by Bregman (1990), sensory feature binding through an oscillatory neural network based on work by Wang (1995; 1996; 1999; 2005; 2008), work by Itti and Koch (2001a) for the saliency map, and finally, work by Wrigley and Brown (2004) for the architecture of single feature processing streams, the inhibition of return of the activation and the attentional leaky integrate and fire neuron. The model was tested using stimuli and an experimental paradigm used by Carlyon, Cusack, Foxton and Robertson (2001). Several modifications were then implemented to the initial model to bring it closer to psychological and cognitive validity. The second part of the thesis furthers the knowledge available concerning the influence of the time spent attending to a task on streaming. Two deviant detection experiments using triplet stimuli are presented. The first experiment is a follow-up of Thompson, Carlyon and Cusack (2011) and replicated their behavioural findings, showing that the time spent attending to a task enhances streaming, and that deviant detection is easier when one stream is perceived. The ERP results showed double decisions markers indicating that subjects may have made their deviant detection based on the absence of the time delayed deviant and confirmed their decision with its later presence. The second experiment investigated the effect of the time spent attending to the task in presence of a continuity illusion on streaming. It was found that the presence of this illusion prevented streaming in such a way that the pattern of the triplet was strengthened through time instead of separated into two streams, and that the deviant detection was easier the longer the subjects attended to the sound sequence.

Yes / This paper investigates the feasibility of using artificial neural networks (NNs) to predict the shear capacity of concrete members reinforced longitudinally with fibre reinforced polymer (FRP) bars, and without any shear reinforcement. An experimental database of 138 test specimens failed in shear is created and used to train and test NNs as well as to assess the accuracy of three existing shear design methods. The created NN predicted to a high level of accuracy the shear capacity of FRP reinforced concrete members. Garson index was employed to identify the relative importance of the influencing parameters on the shear capacity based on the trained NNs weightings. A parametric analysis was also conducted using the trained NN to establish the trend of the main influencing variables on the shear capacity. Many of the assumptions made by the shear design methods are predicted by the NN developed; however, few are inconsistent with the NN predictions.

No / The water permeability constant, (K-w), is one of the many important parameters that affect optimal design and operation of RO processes. In model based studies, e.g. within the RO process model, estimation of W-w is therefore important There are only two available literature correlations for calculating the dynamic K-w values. However, each of them is only applicable for a given membrane type, given feed salinity over a certain operating pressure range. In this work, we develop a time dependent neural network (NN) based correlation to predict K-w in RO desalination processes under fouling conditions. It is found that the NN based correlation can predict the K-w values very closely to those obtained by the existing correlations for the same membrane type, operating pressure range and feed salinity. However, the novel feature of this correlation is that it is able to predict K-w values for any of the two membrane types and for any operating pressure and any feed salinity within a wide range. In addition, for the first time the effect of feed salinity on Kw values at low pressure operation is reported. Whilst developing the correlation, the effect of numbers of hidden layers and neurons in each layer and the transfer functions is also investigated. (C) 2014 Elsevier B.V. All rights reserved.