Deep Neural Networks (DNN) are nothing but neural networks with many hidden layers. DNNs are becoming popular in automatic speech recognition tasks which combines a good acoustic with a language model. Standard feedforward neural networks cannot handle speech data well since they do not have a way to feed information from a later layer back to an earlier layer. Thus, Recurrent Neural Networks (RNNs) have been introduced to take temporal dependencies into account. However, the shortcoming of RNNs is that long-term dependencies due to the vanishing/exploding gradient problem cannot be handled. Therefore, Long Short-Term Memory (LSTM) networks were introduced, which are a special case of RNNs, that takes long-term dependencies in a speech in addition to short-term dependencies into account. Similarily, GRU (Gated Recurrent Unit) networks are an improvement of LSTM networks also taking long-term dependencies into consideration. Thus, in this paper, we evaluate RNN, LSTM, and GRU to compare their performances on a reduced TED-LIUM speech data set. The results show that LSTM achieves the best word error rates, however, the GRU optimization is faster while achieving word error rates close to LSTM.

An intrusion detection system (IDS) is an important feature to employ in order to protect a system against network attacks. An IDS monitors the activity within a network of connected computers as to analyze the activity of intrusive patterns. In the event of an ‘attack’, the system has to respond appropriately. Different machine learning techniques have been applied in the past. These techniques fall either into the clustering or the classification category. In this paper, the classification method is used whereby a neural network ensemble method is employed to classify the different types of attacks. The neural network ensemble method consists of an autoencoder, a deep belief neural network, a deep neural network, and an extreme learning machine. The data used for the investigation is the NSL-KDD data set. In particular, the detection rate and false alarm rate among other measures (confusion matrix, classification accuracy, and AUC) of the implemented neural network ensemble are evaluated.

Internal corrosion is a major concern in ensuring the safety of transmission and gathering pipelines in Structural Health Monitoring (SHM). It usually requires numerous sensors deployed inside the piping system to comprehensively cover the locations with high corrosion rates. This study presents a hybrid modeling strategy using Computational Fluid Dynamics (CFD) and Genetic Algorithm (GA) to improve the sensor placement scheme for corrosion detection and monitoring. The essence of the proposed strategy harnesses the well-validated physical modeling capability of the CFD to simulate the oil-water two-phase flow and the stochastic searching ability of the GA to explore better solutions on a global level. The CFD-based corrosion rate prediction was validated through experimental results and further used to form the initial population for GA optimization. Importantly, fitness was defined by considering both sensing effectiveness and cost of sensor coverage. The hybrid modeling strategy was implemented through case studies, where three typical pipe fittings were used to demonstrate the applicability of the sensor layout design for corrosion detection in pipelines. The GA optimization results show high accuracy for sensor placement inside the pipelines. The best fitness of the U-shaped, upward-inclined, and downward-inclined pipes were 0.9415, 0.9064, and 0.9183, respectively. Upon this, the hybrid modeling strategy can provide a promising tool for the pipeline industry to design the practical placement.