Search

Introduction: The end product of purine metabolism in humans is uric acid (UA). Although uric acid can function as either an antioxidant or an oxidant depending on the surrounding environment, the chemical environment can also impact uric acid.The uric acid level in the serum can predict the development of diabetic nephropathy in type 1 diabetes. Objective: This study aims to determine factors that are perhaps having an association with acid uric. Method: Variables selection is basedon clinical importance. The most significant variable will be assigned and analyzed using Artificial Neural Network (ANN) through multilayer feed-forward and contour plot. Results: Through the architecture of MLFF with two hidden layers, it was found that Creatinine level, Urea level, Systolic Blood Pressure reading, Waist circumference reading, Gender play an essential role toward uric level with an accuracy of 97.7% and the predicted mean squared error (MSE.net) is 0.005. The combination of the selected variable showing the highest significance in predicting the level of uric acid. Conclusion: These findings offer useful future management action plans for patients with diabetes.By controlling these four variables can improve the level of health among diabetic patients. Bangladesh Journal of Medical Science Vol.20(4) 2021 p.741-747

The electrical load forecasting is a fundamental technique for consumer load prediction for utilities. The accurate load forecasting is crucial to design Demand Response (DR) programs in the paradigm of smart grids. Artificial Neural Network (ANN) based techniques have been widely used in recent years and applied to predict the electric load with high accuracy to participate in DR programs for commercial, industrial and residential consumers. This research work is focused on the use and comparison of two ANN-based load forecasting techniques, i.e. Feed-Forward Neural Network (FFNN) and Echo State Network (ESN), on a dataset related to commercial buildings, in view of a possible DR program application. The results of both models are compared based on the load forecasting accuracy through experimental measurements and suitably defined metrics.

Due to the computational complexities of artificial neural networks, MLaaS (machine learning as a service), which is one of the main cloud computing services, is taking the responsibility of the neural network training. With the increase in demand for third-party neural network training, there is a high possibility of adversarial attacks through malicious training. Backdoor attacks are among the most efficient attacks which cause targeted misclassification while the accuracy on clean data is not affected. In this paper, we provide the first investigation about the influence of applying scale-free networks to feed-forward neural networks (FFNNs) against backdoor attacks inserted via the MNIST dataset. It is the first time that the feed-forward neural network structure is changed to improve the network robustness against backdoor attacks using scale-free structure before the network is getting attacked. It has been achieved that scale-free neural networks with long range connections not only keep the accuracy high with strong stability but also make it independent of the number of hidden layers and prevent overfitting.

Feed-forward neural networks have been frequently used in forecasting problems, recently. In this study, we propose a naive method to improve the forecasting ability of feed-forward neural networks with a single hidden layer by adapting meta fuzzy functions. Because neural networks are very sensitive to the initial random weights, usually some numbers of repeats are processed with different initial random weights. The forecasts for the different repeats are, then, averaged with equal weights to obtain more reliable results. However, if we can assign the correct initials with more appropriate weights, then, neural networks can produce very competitive outcomes. In this sense, rather than assigning the equal weights for different repeats with different initials, meta fuzzy functions are used to investigate the best/better forecast with assigning different weights. 4 datasets are used to verify the performance of the proposed method in terms of RMSE and MAPE metrics.

Sound field recording using spherical harmonics (SH) has been widely used. However, too many microphones are needed when recording sound fields over large areas, due to the capture of the higher order of spherical harmonic coefficients. The theory of GO in deep learning inspired us. With training the data much less than all GO’s legal positions data, the Alpha Go has defeated top GO players. According to the information learned from a specific dataset, the higher spherical harmonics coefficients may be estimated with few captured sound pressures. In this paper, a learning-based approach for estimation of the SH coefficients has been investigated. In the proposed approach, SH coefficients are estimated with a feed-forward neural network (FNN) based on measurements of a spherical array. We generate a uniformly distributed dataset, try to evaluate the method on an average situation. Moreover, with the real sound field data in the SOFiA dataset, we try to evaluate the performance of our method when the correlations of data are weak. Experimental results show that the proposed approach achieves higher estimation accuracy of SH coefficients than a previously reported method. In simulations, 9 microphones’ performance using the proposed approach can approximate an array with 16 microphones. The experiments confirmed the feasibility of estimating the SH coefficients with the data-driven method. Thus in a specific application, it can be used to reduce the required number of microphones.

Wireless networks have evolved over the years and they have become some of the most prominent communication media. These networks generally transmit large volumes of information at any given time. This has engendered a number of security threats and privacy concerns. This paper presents a Deep Long Short-Term Memory (DLSTM) based classifier for wireless intrusion detection system (IDS). Using the NSL-KDD dataset, we compare the DLSTM IDS to existing methods such as Deep Feed Forward Neural Networks, Support Vector Machines, k-Nearest Neighbors, Random Forests and Naive Bayes. The experimental results suggest that the DLSTM IDS outperformed existing approaches.

We propose a general framework for finding the ground state of many-body fermionic systems by using feed-forward neural networks. The anticommutation relation for fermions is usually implemented to a variational wave function by the Slater determinant (or Pfaffian), which is a computational bottleneck because of the numerical cost of $O(N^3)$ for $N$ particles. We bypass this bottleneck by explicitly calculating the sign changes associated with particle exchanges in real space and using fully connected neural networks for optimizing the rest parts of the wave function. This reduces the computational cost to $O(N^2)$ or less. We show that the accuracy of the approximation can be improved by optimizing the "variance" of the energy simultaneously with the energy itself. We also find that a reweighting method in Monte Carlo sampling can stabilize the calculation. These improvements can be applied to other approaches based on variational Monte Carlo methods. Moreover, we show that the accuracy can be further improved by using the symmetry of the system, the representative states, and an additional neural network implementing a generalized Gutzwiller-Jastrow factor. We demonstrate the efficiency of the method by applying it to a two-dimensional Hubbard model.

Environmental sound recognition is currently an important and valuable field of computer science and robotics, , security or environmental protection. The underlying methodology evolved from primary speech application characteristic methods to more specific approaches, and with the advent of the deep learning paradigm many attempts using these methods arose. The paper reopens the research we have started on the application of the Feed Forward Neural Networks, by exploring several configurations, and introduces the Convolutional Neural Networks in our investigation. The experiments consider three classes of forest specific sounds and meant to detect the chainsaw sounds chainsaw, vehicle, and genuine forest.

The essence of Machine Learning modelling may be summarized as: to unravel the implicit pattern in the data by having access to only a finite number of samples. The theory which studies this process is rich, and two quantities are of particular importance: the performance errors inside and outside the sample. The error inside the sample is called the training error, and is calculated over the set used to optimize the models parameters. The outside error is the average error amongst all samples, and may be understood as the true error. Although the final objective is to construct a model with low true error, we only have access to the training error, which is an empirical estimation. Thus, in order to deduce the general pattern in the data, is necessary that both are similar.The distance between those errors is called the generalization gap, and much of the theory is dedicated to study its properties and upper bounds, so that we may understand under what circumstances it is controlled. The gap is a measure of the models ability to properly induce the global pattern, and is a major topic in all applied instances of Machine Learning.The classical view of statistics correlates the generalization property with the model capacity to fit patterns in the data. The reasoning behind this is that, being capable of fitting many configurations, the model is prone to read noise in the sample, and thus perform poorly in general. However, the definition of a models complexity is loose, and while it usually translates to the number of parameters, there is one hypothesis space which seemingly escapes this intuition. That is the case of Neural Networks. Using Neural Networks with many layers (Deep learning) is proving to be the best modelling paradigm for many benchmark problems, and many of the advances in the industry are due to their success. However, this seemingly goes against what classical Statistical Learning theory states about generalization and complexity, since Deep networks are capable offitting many patterns. Indeed, there has been experiments showing that networks may fit even random labels.This apparent paradox is an open question in the field and the main topic of this work.After a introduction and overview of the classical understanding of generalization, we introduce the work of [20], which is the central to our contributions. In it a new approach named Analytical Learning is proposed, aiming to complement the classical one, hopefully bringingsome insights about the apparent contradiction emerged from Deep Learning. Instead of analyzing probabilistic bounds, in this paper the generalization gap is studied in a context where the predictor and the dataset are fixed. By doing so, we prevent the pessimistic cases, and a tighter bound is hopefully achieved. Additionally, it provides a more real scenario, since in practice usually the data is given. The main result of [20] bounds the gap involving a term related to the data and another related to the loss function\'s Hardy-Kruase Variation. Our main contribution revolves around tracing similarities between this variation term with the stability concept studied in the classical approach of generalization, making parallels with what may be understood in the Analytical case as information.The main idea is that the loss function variation decreases if the partial derivatives of the predictor, according to the instance space, are close to the oracles. The derivatives in this sense may be understood as how much information the function is reading, since it measures the impact of a certain dimension in a local prediction. Thus, if the predictor reads information similarly to the oracle, then we guarantee a low gap. With this, we argue that the partial derivatives of the predictor are the main measure of regularization in the analytical sense. One of the advantages of this is simplicity: rewriting the SGD (Stochastic Gradient Descent) optimization step in the function space, we have an easy way to investigate the evolution of the models complexity during training.Furthermore, we use this interpretation to develop on relative recent papers trying to tackle the generalization paradox in deep learning, [28] and [37]. In the former we make an extensive analysis. while in the latter we make a more brief qualitative approach, showing how our interpretation relates with their result. In [28], the complexity of networks is studied through the lens of Fourier Theory. There is shown that the space of ReLU (Rectifier Linear Unit) networks has a high spectral decay:during optimization, the increments in the k-th harmonic caused by the weights updates decreases with at least k^2. This means that high frequencies magnitudes in this space are naturally damped during training, suggesting an inherent regularization property. However,at no moment in [28] the generalization gap is mentioned, and so it is not clear if the spectral decay is enough to guarantee a good estimation of the true error. Motivated by this, we show a bound using the Hardy-Krause Variation on splines which decreases with the degree, justifying the special properties of ReLU activation functions.In [37] the main theorem shows that if the architecture of the network follows a funnel pattern (when the number of neurons in the network decreases as we go deeper), then increasing the number of layers actually reduces the generalization gap, thus supporting the deep learning approach. This happens because the funnel like architecture forces a nontrivial kernel in the linear transformations, which translates into a loss of information. This implies that as the number of layers increases, the information shared between the final layerand the dataset decreases, making the prediction less data dependant and thus regularized.This result relates closely to our interpretation of information in the analytical sense.Having a non trivial kernel means that in some cases the prediction is constant with respect to disruptions in certain dimensions. This means that the overall variation (in the sense ofderivatives) will be smaller, which according to Analytical Learning translates to a smaller generalization gap. A essência do Aprendizado de Máquina pode ser resumida como: revelar padrões implícitos nos dados tendo em posse apenas uma amostra finita. A Teoria que estuda essa questão é vasta, e duas quantidades são de particular importância: os erros do modelo dentro e fora da amostra. O primeiro é chamado de erro de treinamento, e mede a performance dentro do conjunto usado para otimizar. O segundo é a média do erro em todas as amostras, e pode ser compreendido como o erro real. Apesar do verdadeiro indicador de performance ser o erro real, podemos calcular apenas a performance dentro da amostra, que é uma estimativa empírica. Portanto, para deduzir o padrão geral, é necessário que tais erros sejam similares. A distância entre estes erros é chamada de generalization gap, e grande parte da teoria se dedica a estudar as suas propriedades e limites superiores. Ele é uma medida da habilidade do modelo de induzir corretamente o comportamento global, e é um tópico central em todas as aplicações de Aprendizagem de Máquina. A visão clássica da Estatística correlaciona a propriedade de generalização com a capacidade do modelo de ajustar os dados. A ideia é de que, se o modelo é capaz de performar em diversas configurações, então ele será sensível a ruídos presentes, e então performará mal fora da amostra. Entretanto, a definição de complexidade de um modelo é vaga, e apesar de muitas vezes ser caracterizado pelo número de parâmetros, existe um espaço de hipóteses que aparentemente escapa essa intuição. É o caso das redes neurais profundas. Fazer uso de redes com muitas camadas (aprendizado profundo) está provando ser um dos melhores paradigmas de modelagem em diversos problemas de referência, e os avanços da indústria se devem em grande parte ao seu grande sucesso. Entretanto, isso aparentemente contradiz o que a teoria de Aprendizado Estatístico nos fala sobre complexidade e generalização, já que redes profundas são capazes de ajustar diversos padrões. De fato, existem experimentos que mostram que elas podem ajustar até variáveis respostas aleatórias. Esse possível paradoxo é uma questão em aberto na área e é o principal tópico neste estudo. Após uma introdução e revisão sobre a teoria clássica de generalização, nós introduzir o trabalho de [20], que é central para as nossas contribuições. Nele é proposta uma nova abordagem chamada de Analytical Learning com o intuito de complementar o entendimento clássico, almejando trazer uma nova visão sobre essa aparente contradição revelada pelo aprendizado profundo. Neste artigo, ao invés de analisar limites de natureza probabilística, o generalization gap é estudado em um contexto onde o preditor e o conjunto de dados é fixo. Assim, desconsideramos os casos pessimistas, e uma majoração mais precisa poderia ser alcançada. Além disso, essa proposta considera um cenário mais real, pois na prática a amostra é dada. O principal resultado de [20] é uma majoração que envolve um termo relacionado aos dados, e um outro relacionado à Variação de Hardy-Krause da função perda. Nossa principal contribuição envolve traçar similaridades entre esse termo de variação com a noção de estabilidade estudada na abordagem estatística da generalização, fazendo paralelos com o que pode ser entendido no caso Analítico como informação. A ideia principal é que a variação da função perda decresce se as derivadas parciais do preditor são próximas das do oráculo. Dessa forma, as derivadas nesse sentido podem ser compreendidas como a quantidade de informação que está sendo usada, já que ela mede o impacto de uma certa dimensão na predição local. Portanto, se o preditor lê a informação da mesma forma que a função que gerou os dados, então podemos garantir uma aproximação do erro de treinamento com o erro real. Com isso, argumentamos que a derivada parcial é a principal forma de medir a regularização de um modelo no sentido analítico. Uma das vantagens dessa abordagem é simplicidade: reescrevendo o passo de otimização do DGE (descida pelo gradiente estocástico) no espaço das funções, adquirimos uma forma simples de investigar a evolução da complexidade do modelo durante o treinamento. Ademais, usamos essa interpretação para elaborar em recentes artigos abordando o problema da generalização em aprendizado profundo, [28] e [37]. No primeiro fazemos uma análise extensiva, enquanto no segundo fazemos uma análise qualitativa mais abreviada, mostrando como nossa interpretação se relaciona com os resultados neles presentes. Em [28], a complexidade das redes neurais é estudada sob a lente da Teoria de Fourier. Nele é mostrado que o espaço de redes ReLU (Rectifier Linear Unit) apresenta um decaimento espectral particularmente intenso: durante a otimização, os incrementos no k-ésimo harmônico decorrente da atualização dos pesos decresce com pelo menos k². Isso significa que as magnitudes de frequências mais altas são naturalmente amortecidas durante o treinamento, sugerindo uma propriedade regularizadora intrínseca no espaço. Apesar disso, em momento algum [28] menciona o generalization gap, e portanto não é claro se o decaimento espectral é suficiente para garantir que o erro de treinamento seja uma boa estimativa do erro real. Motivados por isso, mostramos um limite superior usando a Variação de Hardy-Krause em splines que decresce com o grau, justificando as propriedades especiais da função de ligação ReLU. O principal teorema em [37] mostra que, para arquiteturas com um formato de funil (quando o número de neurônios decresce à medida que nos aprofundamos na rede), aumentar o número de camadas implica em uma redução no generalization gap, fundamentando portanto o aprendizado profundo nesses casos. Isso ocorre pois uma arquitetura com tal formato força núcleos não triviais nas transformações lineares, o que implica em uma perda de informação. Ou seja, ao aumentar o número de camadas, a quantidade de informação nos dados usada pelo preditor diminui, tornando a predição menos dependente da amostra, e portanto regularizada. Esse resultado se relaciona muito com a nossa interpretação de informação no sentido analítico. Ter um núcleo não trivial na arquitetura significa que em alguns casos a predição vai permanecer inalterada com relação à mudanças em certas dimensões. Isso implica que a variação geral (no sentido de derivadas) será menor, o que de acordo com a teoria do Aprendizado Analítico, acarreta em uma melhor estimativa do erro real.

Liquidity plays a vital role in the financial markets, affecting a myriad of factors including stock prices, returns, and risk. In the stock market, liquidity is usually measured through the order book, which captures the orders placed by traders to buy and sell stocks at different price points. The introduction of electronic trading systems in recent years made the deeper layers of the order book more accessible to traders and thus of greater interest to researchers. This paper examines the efficacy of leveraging the deeper layers of the order book when forecasting quoted depth—a measure of liquidity—on a per-minute basis. Using Deep Feed Forward Neural Networks, we show that the deeper layers do provide additional information compared to the upper layers alone.

By studying the effects of different types of feed forward neural networks in different fields, the applicable environment of different neural networks can be judged, which will make it easier for people to choose appropriate neural network when it's needed. To achieve this, in this article I summarize and classify the existing neural network experiences and feedback results, and compare the data before and after using the neural network. The data shows that BP networks can improve the resolution or accuracy of problems with no obvious influencing factors. Convolutional networks can increase the accuracy of image processing to more than 95%, and RBF networks can calculate high-precision data curves. Thus, it can be concluded that the BP network is suitable for solving problems with unclear influencing factors, the convolutional network has more image processing problems, and the RBF network has a higher frequency of use when higher results are required.

The article discusses the prerequisites for the use of artificial neural networks with direct connection for diagnosing the state of rotary machines with liquid friction bearings. During the study, an experimental setup with a multi-sensor measuring system was developed. The measuring system is based on sensors that measure different types of data. An experiment was conducted in which various types of failures were modeled. Data was processed to provide their network. Various ways of presenting data were used. The effect of various types of measurements on the accuracy of networks was investigated. The principles of preparing data sets are presented. A study was conducted of the influence of the volume of the training sample on the accuracy of the artificial neural network. The possibilities of using multisensory measurements of various nature for the diagnosis of rotary machines are shown.

Deep Learning (DL) is a research area that has flourished significantly in recent years and has shown remarkable potential for artificial intelligence in the field of medical applications. We have implemented the DL algorithm for the diabetes classification. This paper applied the Multi-Layer Feed Forward Neural Networks (MLFNN) for the diabetes classification on the Pima Indian Diabetes datasets. Furthermore, various activation functions, learning algorithms, and techniques to handle missing values are considered to enhance the classification accuracy of the diabetes dataset. Finally, the outcomes of experiments are compared with two machine learning algorithms like Nave Bayes and Random Forest. The achieved classification accuracy by MLFNN (84.17%) is the best of all the other classifiers.

This paper focuses on reactive power optimization based on deep reinforcement learning. A deep reinforcement learning algorithm DQN is selected as an improvement of the reinforcement learning algorithm, this algorithm uses a deep neural network instead of a Q value table in the reinforcement learning algorithm, the method of power flow calculation in the Matlab toolbox is executed to verify the optimization effect. The single-section training results of the 14-BUS IEEE power grid model are tested and the parameters of the power grid model are modified after the optimization is confirmed. The DQN model is trained in a simulated complex environment, and the performance of the optimization effect is tested after the training is completed. Reactive power optimization of the power grid based on deep reinforcement learning is validated by experimental results.

Objectives We aimed to evaluate the ability of feed-forward neural networks (fNNs) to predict the neurodevelopmental outcome (NDO) of very preterm neonates (VPIs) at 12 months corrected age by using biomarkers of cerebral MR proton spectroscopy (1H-MRS) and diffusion tensor imaging (DTI) at term-equivalent age (TEA). Methods In this prospective study, 300 VPIs born before 32 gestational weeks received an MRI scan at TEA between September 2013 and December 2017. Due to missing or poor-quality spectroscopy data and missing neurodevelopmental tests, 173 VPIs were excluded. Data sets consisting of 103 and 115 VPIs were considered for prediction of motor and cognitive developmental delay, respectively. Five metabolite ratios and two DTI characteristics in six different areas of the brain were evaluated. A feature selection algorithm was developed for receiving a subset of characteristics prevalent for the VPIs with a developmental delay. Finally, the predictors were constructed employing multiple fNNs and fourfold cross-validation. Results By employing the constructed fNN predictors, we were able to predict cognitive delays of VPIs with 85.7% sensitivity, 100% specificity, 100% positive predictive value (PPV) and 99.1% negative predictive value (NPV). For the prediction of motor delay, we achieved a sensitivity of 76.9%, a specificity of 98.9%, a PPV of 90.9% and an NPV of 96.7%. Conclusion FNNs might be able to predict motor and cognitive development of VPIs at 12 months corrected age when employing biomarkers of cerebral 1H-MRS and DTI quantified at TEA. Key Points • A feed-forward neuronal network is a promising tool for outcome prediction in premature infants. • Cerebral proton magnetic resonance spectroscopy and diffusion tensor imaging can be used for the construction of early prognostic biomarkers. • Premature infants that would most benefit from early intervention services can be spotted at the time of optimal neuroplasticity.

The world is moving to an autonomous era. Autonomous vehicles take a major role in day-to-day activity, helping human personnel do work quickly and independently. Unmanned aerial vehicles (UAVs) are autonomous vehicles controlled using remotes in ground station by human personnel. These UAVs act as a network that plays a vital role in the digital era. There are different architectures of UAV networks available. This chapter concentrates on centralized UAV network. Because of wireless and autonomy characteristics, these networks are prone to various security issues, so it's very important to monitor and analyze the traffic of the UAV network in order to identify the intrusions. This chapter proposes enhanced deep feed forward neural network (EDFFNN) in order to monitor and analyze the traffic of the UAV network to detect the intrusions with maximum detection rate of 94.4%. The results have been compared with the previous method of intrusion detection.

Deep learning has proven to be an effective tool in predictive analytics due to its ability to understand patterns in large data sets. Supervised Deep learning techniques are utilized to understand the relationships between electric motor aging and the sensor parameters from the same. This approach overcomes the challenge of the requirements of high quality labeled data which are both time consuming and difficult to acquire. Thus, use of supervised deep learning methods has the potential to deploy RUL models with minimal training data. This paper focuses on the implementation of deep learning methods for Remaining Useful Life prediction for electric motor. Keywords—Deep learning, Remaining Useful Life, RUL, Feed Forward Neural Networks

There are research results of artificial neural networks usage for solving a hardly formalized objective – detection of a DDoS attacks on the computer network information resource in this article. An analysis of existing methods, technologies and tools for detecting DDoS attacks and protecting from them is carried out. Several feed forward neural networks are simulated. The architecture of the neural network which provides high-precision detection is presented.

Modeling WiFi RSSI in buildings is crucial for fingerprinting based indoor positioning algorithms. Fingerprinting is a popular approach which converts the localization task into data mining and optimization problems. Although fingerprinting algorithms may rely on any kind of measurements, WiFi RSSI values are the most commonly used due to their ubiquity. However there are existing solutions for modeling WiFi signal propagation which requires knowledge about the environment and could be difficult to integrate into a positioning algorithm. On the other hand, feed forward artificial neural networks could be used to model the WiFi RSSI values easily based on the fingerprinting database. Therefore, training the networks efficiently is important as is the analysis of training parameters. Experiments over the Miskolc IIS Hybrid IPS Dataset showed that networks with ReLu or tanh activation functions and trained by sgd or adam solvers could achieve high accuracy.

Focusing on non-iterative approaches in training feed-forward neural networks, this special issue includes 12 papers to share the latest progress, current challenges, and potential applications of this topic. This editorial presents a background of the special issue and a brief introduction to the 12 contributions.

The ocean is full of a variety of sounds from natural, biological or anthropogenic sources. Listening to the animals sounds allows scientists to detect, identify, and locate different endangered species as well as listening to high intensity anthropogenic sources, which could harm the marine ecosystem. In this work, a new computational model for marine mammals classification is presented and validated with data from an online database. The feature extraction is performed using 1/6 octave analysis and the classification is carried out based on an independent ensemble methodology, where the outputs of four parallel feed forward neural networks are combined to classify eleven possible classes (seven marine mammals plus four additional classes). Unlike similar works, this paper considers multiple sounds emitted by each species such as whistles, calls and squeaks. The model demonstrated favorable performance reaching a classification rate of 90% at a low computational cost.

Extreme learning machine (ELM) is an interesting algorithm for learning the hidden layer of single layer feed forward neural networks. However, one of the main shortcomings restricting further improvement of ELM is the complexity of singular value decomposition (SVD) for computing the Moore-Penrose generalized inverse of the hidden layer matrix. This paper presents a new algorithm named fast adaptive shrinkage/thresholding algorithm ELM (FASTA-ELM) which uses an extension of forward-backward splitting (FBS) to compute the smallest norm of the output weights in ELM. The proposed FASTA-ELM algorithm is evaluated on face gender recognition problem using 5 benchmarked datasets. The results indicate that FASTA-ELM provides efficient performance and outperforms the standard ELM and two other variants of ELM in terms of generalization ability and computational time. Furthermore, the recognition performance of FASTA-ELM is comparable to other state-of-the-art face gender recognition methods.

This chapter introduces the general features of artificial neural networks. After a presentation of the mathematical neural cell, we focus on feed-forward networks. First, we discuss the preprocessing of data and next we present a survey of the different methods for calibrating such networks. Finally, we apply the theory to an insurance data set and compare the predictive power of neural networks and generalized linear models.

Artificial neural networks (ANNs) have shown efficient results in statistics and computer science applications. Feed-forward neural network (FNN) is the most popular and simplest neural network architecture, capable of solving nonlinearity. In this paper, feed-forward neural networks’ weight and bias figuring using a newly proposed metaheuristic Salp Swarm Algorithm (SSA) are proposed. SSA is a swarm-based metaheuristic inspired by the navigating and foraging behaviour of salp swarm. The performance is evaluated for some of the benchmarked datasets and compared with some well-known metaheuristics.

Feed-forward neural networks are efficient at solving various types of problems. However, finding efficient training algorithms for feed-forward neural networks is challenging. The dynamic group optimisation (DGO) algorithm is a recently proposed half-swarm half-evolutionary algorithm, which exhibits a rapid convergence rate and good performance in searching and avoiding local optima. In this paper, we propose a new hybrid algorithm, FNNDGO that integrates the DGO algorithm into a feed-forward neural network. DGO plays an optimisation role in training the neural network, by tuning parameters to their optimal values and configuring the structure of feed-forward neural networks. The performance of the proposed algorithm was determined by comparing its performance with those of other training methods in solving two types of problems. The experimental results show that our proposed algorithm exhibits promising performance for solving real-world problems.

While there are methods to estimate the amount of evaporation, there were no mentions in recent studies of predicting it. In this paper, evaporation in a water reservoir for the succeeding 24 hours is predicted using a feed forward artificial neural network and backpropagation. The neural network is trained using historical data with temperature, humidity, wind speed, and solar radiation as input parameters. Various configurations of the neural network is considered and the best architecture is selected accordingly. The same weather parameters used in the training are gathered using sensors and are used as inputs to predict of the evaporation over the succeeding 24 hours. The predicted or the experimental value of evaporation is then compared to the evaporation estimated using a class A pan. The number of neurons in the hidden layer is varied from two to twelve and is evaluated by mean squared error. 4-7-1, the hidden layer with seven neurons is selected. In testing the accuracy in predicting the evaporation, it resulted to an average error of 4.48%.

Forecasting of wave conditions plays an essential role for offshore construction and maintenance. Recently, machine learning-based wave forecasting models have been developed and their integrated usage with physics-based numerical models has become popular. These studies mostly apply Feed Forward Neural Networks (FFNNs) with an emphasis on prediction of time-series of waves, tides and storm surges. As a particularly different approach, we develop a deep learning-based wave forecasting model using Long Short-Term Memory (LSTM) network under Recurrent Neural Networks. As a case study, the model will be utilized to predict the wave conditions (low or high) near the Tottori Port, Japan.Recorded Presentation from the vICCE (YouTube Link): https://youtu.be/oMvIS9zkIOs

Abundance and biological diversity of phytoplankton communities, investigated in this work, are often used as a marker for the determination of environmental health and fresh water quality. Presently their routine analysis is very time consuming and expensive. A lot of articles are devoted to the development of a system for an in situ automated analysis of phytoplankton properties. However, the applied problems of biology, ecology and, in particular, algology usually are associated with some difficulties due to shortage and/or fuzziness of the experimental data. Hence, while using neural network modelling a set of specific problems can occur. In this article on the base of experimental data several such problems are presented and possible solutions are suggested. In particular, the illogical behavior of classifying neural network is revealed, while studying the biological diversity of cyanobacteria, and the original technique for results validation is presented. This problem is investigated on a set of spectroscopic data, recorded by means of confocal laser scanning microscopy. The generalization quality of the trained model is studied as the main learning parameter. Another problem of shortage dataset is examined in the frames of regression model for bioplankton abundance. This problem is solved by means of feed-forward back-propagation neural networks with two hidden layers. The modelling was carried out on a small experimental selection (only 39 observations were available), despite this, the relatively high determination coefficient was obtained for the training and test samples, while using dropout layout.

Because of gait sequences are naturally three-dimensional data, there have been several tensorial feature extraction methods to deal with tensors while there are effective tensorial classifiers. In this work, by using a linear tensor projection, a new classifier based on neural networks with random weights is introduced. Due to the proposed algorithm can classify gait samples directly without vectorizing them, the intrinsic structure information of the input data can be reserved. In addition, discriminative features sets are generated using MPCA to ascertain classification accuracy. Finally, Extensive experiments are carried out on two gait databases and results are compared against state-of-the-art techniques. It is demonstrated that the proposed algorithm MPCA plus TNNRE achieves better recognition performance.

We aim to estimate concrete compressive strength by using the physical properties of the aggregates that are the main components of concrete. For this aim, concrete samples were prepared with aggregates having different origins and characteristics obtained from different 10 locations of the Marmara region. The compressive strength's results obtained by changing the aggregates were compared by ensuring the other components forming the concrete remained constant. 330 separate experiments were conducted to determine the physical characteristics of the aggregates, and these characteristics were used as input data in a multi layer Feed Forward Network model. The compressive strength of 7 and 28 days of the concrete obtained from the experiments were used as output in to a Multi-layer Feed Forward model. The training and test results from the models coincided closely with experimental results; also, the results were compared to the estimations made with a linear regression method.

Botnet have turned into the most serious security dangers on the present Internet framework. A botnet is most extensive and regularly happens in today's cyber-attacks, bringing about the serious risk of our system resources and association's properties. Botnets are accumulations of compromised computers (Bots) which are remotely regulated by its creator (BotMaster) under a typical Command-and-Control (C&C) framework. Botnets cannot just be implemented utilizing existing well-known applications and additionally developed by unknown or inventive applications. This makes the botnet detection a challenging issue. In this paper proposed an anomaly detection model based on genetic neural network system, which joined the significant global searching capability of genetic algorithm with the precise local searching element of back propagation feed forward neural networks to improve the initial weights of neural network.

This paper presents a neural network-based intrusion detection method for the attacks on a computer network. Neural networks are used to predict unusual activities in the system. In particular, feed forward neural networks with the back propagation training algorithm were employed in this study. We propose a method of intrusion detection based on a combination of genetic algorithm (GA) and multilayer perceptron (MLP) neural network to develop a model for intrusion detection system. All tests were realised with the kdd99 data set. The performance of the proposed method of intrusion detection was evaluated on all KDD99 data set, 10% of the KDD99 data set were used for training the GA-MLP model. This system achieves a top accuracy of up to 93.05%.

The aim of this work is to solve object classification problem using Neural Networks. Two types of neural networks are used and compared, namely, classical Feed Forward Neural Networks and Deep Convolutional Networks. Success of the designed networks is investigated using CIFAR-10 dataset. © 2018 IEEE. 6th International Conference on Control Engineering and Information Technology, CEIT 2018 -- 25 October 2018 through 27 October 2018 -- -- 149175

Recent advancements in the field of Artificial Intelligence can support the post-processing of thermographic data, efficiently, especially for nonlinear or complex thermography scanning routines. This study proposes the implementation of an autonomous/intelligent post-processor that is capable of automatically detecting defects from given thermograms via a Convolutional Neural Networks (CNN) coding, in tandem with a Deep Feed Forward Neural Networks (DFF-NN) algorithm to estimate the defect depth as well. Thus, the proposed NN combination will process (detect and quantify) the defects from acquired thermograms in real-time, and without any human (inspector) intervention. The study shows that employing a pre-trained network, using a relatively small dataset of thermograms for training, can detect and quantify defects in thermographic sequences. In this paper, pre-trained networks with CIFAR-10 and ImageNet databases are used, and followed by a fine-tuning step of the later layers in the network; using a relatively small thermograms dataset. This text will also provide several in-depth studies to compare how transfer learning, state of the art object detection architectures, and the convolutional neural networks influence the performance of the trained post-processing system. The proposed post-processor applied to thermograms obtained from a pulsed-thermography setup testing a Carbon Fiber Reinforced Polymer (CFRP) sample with artificially created sub-surface defects validates the CNN approach.

We present a framework to derive upper bounds on the number of regions that feed-forward neural networks with ReLU activation functions are affine linear on. It is based on an inductive analysis that keeps track of the number of such regions per dimensionality of their images within the layers. More precisely, the information about the number regions per dimensionality is pushed through the layers starting with one region of the input dimension of the neural network and using a recursion based on an analysis of how many regions per output dimensionality a subsequent layer with a certain width can induce on an input region with a given dimensionality. The final bound on the number of regions depends on the number and widths of the layers of the neural network and on some additional parameters that were used for the recursion. It is stated in terms of the $L1$ -norm of the last column of a product of matrices and provides a unifying treatment of several previously known bounds: Depending on the choice of the recursion parameters that determine these matrices, it is possible to obtain the bounds from Montufar et al. [1] (2014), Montufar [2] (2017), and Serra et al. [3] (2017) as special cases. For the latter, which is the strongest of these bounds, the formulation in terms of matrices provides new insight. In particular, by using explicit formulas for a Jordan-like decomposition of the involved matrices, we achieve new tighter results for the asymptotic setting, where the number of layers of the same fixed width tends to infinity.

The number and types of inputs of the prediction system is crucial to a load forecasting task. In many cases, the number of inputs is increased, a fact that may lead to poor forecasting accuracy. The scope of the present paper is to examine the effect of dimensionality reduction methods in the forecasting performance of Feed Forward Neural Networks (FFNN). The test case refers to the day-ahead forecasting of the loads connected to a high-to-medium voltage transformer of the transmission system of Greece.

[EN] In this paper, an optimal model is developed for path loss predictions using the Feed-Forward Neural Network (FFNN) algorithm. Drive test measurements were carried out in Canaanland Ota, Nigeria and Ilorin, Nigeria to obtain path loss data at varying distances from 11 different 1,800 MHz base station transmitters. Single-layered FFNNs were trained with normalized terrain profile data (longitude, latitude, elevation, altitude, clutter height) and normalized distances to produce the corresponding path loss values based on the Levenberg-Marquardt algorithm. The number of neurons in the hidden layer was varied (1-50) to determine the Artificial Neural Network (ANN) model with the best prediction accuracy. The performance of the ANN models was evaluated based on different metrics: Mean Absolute error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), standard deviation, and regression coefficient (R). Results of the machine learning processes show that the FNN architecture adopting a tangent activation function and 48 hidden neurons produced the least prediction error, with MAE, MSE, RMSE, standard deviation, and R values of 4.21 dB, 30.99 dB, 5.56 dB, 5.56 dB, and 0.89, respectively. Regarding generalization ability, the predictions of the optimal ANN model yielded MAE, MSE, RMSE, standard deviation, and R values of 4.74 dB, 39.38 dB, 6.27 dB, 6.27 dB, and 0.86, respectively, when tested with new data not previously included in the training process. Compared to the Hata, COST 231, ECC-33, and Egli models, the developed ANN model performed better in terms of prediction accuracy and generalization ability., This work was supported by Covenant University [grant number CUCRID-SMARTCU-000343].

Artificial Neural Network (ANN) provides an attractive alternative tool for researchers for agricultural forecasting. The Multi Layer Feed Forward Neural Net (MLFFNN) is one of the most widely used neural nets. Here, the MLFFNN architecture is examined and compared with time series models such as Autoregressive Integrated Moving Average (ARIMA) model for prediction of agricultural production. Both models were compared using visualization technique and statistical tests and the results were illustrated numerically and graphically.