Search

This paper presents a study of the resistive behavior of a Shape Memory Alloy spring, with a focus on the application of electrical resistance feedback in control systems. Artificial Neural Networks of different topologies were designed to learn the relation between spring electrical resistance and the force exerted. The feedback between layers in Neural Networks is demonstrated to be a key parameter in learning the non-linear and hysteretic behavior of Shape Memory Alloys. Experiments with closed-loop systems showed that shape memory alloy springs generated forces that converged satisfactorily to the desired reference values. The scientific contribution of this work is the use of electrical resistance variation as feedback for controlling the spring force, eliminating the use of an external force sensor. Neural networks were used for both, the sensing process and the system control; in that way the nonlinear and hysterical behavior of the shape memory alloy actuator was well considered. [ABSTRACT FROM AUTHOR]

This paper presents a heterogeneous ensemble classifier for price trend prediction of a stock, in which the prediction results are subsequently used in trading recommendation. The proposed ensemble model is based on Support vector machine, Artificial neural networks, Random forest, Extreme gradient boosting, and Light gradient boosting machine. A feature selection is performed to choose an optimal set of 45 technical indicators as input attributes of the model. Each base classifier is executed with an extensive hyperparameter tuning to improve performance. The prediction results from five base classifiers are aggregated through a modified majority voting among three classifiers with the highest accuracies, to obtain final prediction result. The performance of proposed ensemble classifier is evaluated using daily historical prices of 20 stocks from Stock Exchange of Thailand, with 3 overlapping datasets of 5-year intervals during 2014-2020 for different market conditions. The experimental results show that the proposed ensemble classifier clearly outperforms buy-and-hold strategy, individual base classifiers, and the ensemble with straightforward majority voting in terms of both trading return and Sharpe ratio. [ABSTRACT FROM AUTHOR]

In recent years, the rates of firearm-related violent acts worldwide have risen significantly. It is a severe social problem that compromises the safety of every individual to some extent. This situation has motivated researchers to find new ways to improve the current state-of-the-art solutions, such as automatic surveillance systems, to detect and classify the presence of firearms within a specific scene. These systems reduce the drawbacks of using direct human supervision. Among the available solutions for the classification task, the performance of Deep Learning models stands out, especially those based on Convolutional Neural Networks. Since they start learning directly from raw data (e.g., images), their learning process can be improved even further by using Transfer Learning techniques. However, the classification accuracy depends significantly on choosing the optimum set of values for the different hyper-parameters composing them. Thus, this paper analyses the improvement in the performance of an image-based handgun classification algorithm when tuning its hypermeters values instead of using its default values. In this work, we evaluated the performance variation using two benchmarks Convolutional Neural Networks architectures: AlexNet and Inception V3. We obtained a maximum accuracy of 94.11% when using the Inception V3 network and transfer learning. We employed Nadam as the optimizer and a learning rate equal to 0.0001, a batch size equal 256, and a total of 13 epochs. Experimental results suggest an essential relationship between the performance of the classification model and the data set, the specific combinations of values for the selected optimizer, the batch size, and the learning rate. The obtained improvement in the accuracy was up to 10.33% after the tuning process. [ABSTRACT FROM AUTHOR]

Artificial Neural Networks (ANNs) models play an increasingly significant role in accurate time series prediction tools. However, an accurate time series forecasting using ANN requires an optimal model. Hence, great forecasting methods have been developed from optimized ANN models. Most of them focus more on input variables selection and preprocessing, topologies selection, optimum configuration and its associated parameters regardless of their input variables disposition. This paper provides an investigation of the effects of input variables disposition on ANNs models on training and forecasting performances. After investigation, a new ANNs optimization approach is proposed, consisting of finding optimal input variables disposition from the possible combinations. Therefore, a modified Back-Propagation neural networks training algorithm is presented in this paper. This proposed approach is applied to optimize the feed-forward and recurrent neural networks architectures; both built using traditional techniques, and pursuing to forecast the wind speed. Furthermore, the proposed approach is tested in a collaborative optimization method with single-objective optimization technique. Thus, Genetic Algorithm Back-Propagation neural networks aim to improve the forecasting accuracy relative to traditional methods was proposed. The experiment results demonstrate the requirement to take into consideration the input variables disposition to build a more optimal ANN model. They reveal that each proposed model is superior to its old considered model in terms of forecasting accuracy and thus show that the proposed optimization approach can be useful for time series forecasting accuracy improvement. [ABSTRACT FROM AUTHOR]

Crop yield prediction is an important aspect of agriculture. The timely and accurate crop yield predictions can be of great help for policy makers and farmers in planning and decision making. Generally, statistical models are employed to predict the crop yield which is time consuming and tedious. Emerging trends of deep learning and machine learning has come up as a major breakthrough in the arena. Deep learning models have the inherent ability to perform feature extraction in large dataset thus more suitable for predictions. In this paper, a deep learning-based Recurrent Neural Network (RNN) model is employed to predict wheat crop yield of northern region of India. The present study also employed LSTM to unravel the vanishing gradient problem inherent in RNN model. Experiments were conducted using 43 years benchmark dataset and proposed model results were compared with three machine learning models. Evidently, the results obtained from RNN-LSTM model(RMSE: 147.12,MAE: 60.50), Artificial Neural Network(RMSE: 732.14,MAE: 623.13), Random Forest (RMSE: 540.88, MAE: 449.36) and Multivariate Linear Regression (RMSE: 915.64,MAE: 796.07), proved the efficacy of model. Also, predicted crop yield values were found to be more close to true values for RNN-LSTM model proving efficiency of the proposed work. [ABSTRACT FROM AUTHOR]

Finding the best structure of ANN to minimize errors, the processing and the search time is one of the main objectives in the AI field. In order to achieve prediction with a high degree of accuracy in a short time, an enhanced PSO-based selection technique to determine the optimal configuration for the artificial neural network has been proposed in this paper. To design the neural network to minimize processing time, search time and maximize the accuracy of prediction, it is necessary to identify hyperparameter values with precision. PSO with 2-D search space has been employed to select the best hyperparameters in order to construct the best neural network where PSO is used as a decision-making model and ANN is used as a learning model. The suggested technique was used to select the optimal number of the hidden layer and the number of units per hidden layer. The proposed technique was evaluated using a chemical dataset. The result of testing the proposed technique displayed high prediction accuracy with MSE equal to 3.9% and the relative error between the expected output and actual target is less than 1.6%. The results of the comparison of the proposed technique with the ANN showed thatthe proposed approach could predict output with an infinitesimal error, outperforming the existing ANN model in terms of error ratio. [ABSTRACT FROM AUTHOR]

Artificial intelligence (AI) has been widely affecting our lives in many ways for the last ten years. Potential applications of AI are currently being used in many sectors. However, the usage of AI has been very limited in sports science compared to the other sectors. In this study, we developed an artificial neural network model to estimate the market values of the players in the Turkish Super Football League. While the market value was selected as an output, the input values were: minutes played; goals scored; xG: assists; xA; defensive duels won %; tackle success %; shots on target %; Short-middle pass accuracy %; long pass accuracy %; and accuracy of passes to penalty area. After creating a neural network based on the input-output values with high training, validation and testing statistical values, input values were computed with the neural network created and then the output values were estimated. In conclusion, an artificial neural network is becoming one of the important modeling methods in all areas of life. Although the application of artificial neural networks is very limited in sports science, it is one of the suitable science disciplines where there is a lot of statistical data. The methodology proposed in this paper can also be used for talent selection in football. Moreover, it may help stop criticism by TV sports programmes and newspapers because market value estimation is based on fair performance parameters. Further studies are strongly recommended, not only for football but also for other sports disciplines. [ABSTRACT FROM AUTHOR]

Scene recognition is used in many computer vision and related applications, including information retrieval, robotics, real-time monitoring, and event-classification. Due to the complex nature of the task of scene recognition, it has been greatly improved by deep learning architectures that can be trained by utilizing large and comprehensive datasets. This paper presents a scene classification method in which local and global features are used and are concatenated with the DCT-Convolutional features of AlexNet. The features are fed into AlexNet's fully connected layers for classification. The local and global features are made efficient by selecting the correct size of Bag of Visual Words (BOVW) and feature selection techniques, which are evaluated in the experimentation section. We used AlexNet with the modification of adding additional dense fully connected layers and compared its result with the model previously trained on the Places365 dataset. Our model is also compared with other scene recognition methods, and it clearly outperforms in terms of accuracy. [ABSTRACT FROM AUTHOR]

The information resulting from the use of the organization's products and services is a valuable resource for business analytics. Therefore, it is necessary to have systems to analyze customer reviews. This article is about categorizing and predicting customer sentiments. In this article, a new framework for categorizing and predicting customer sentiments was proposed. The customer reviews were collected from an international hotel. In the next step, the customer reviews processed, and then entered into various machine learning algorithms. The algorithms used in this paper were support vector machine (SVM), artificial neural network (ANN), naive bayes (NB), decision tree (DT), C4.5 and k-nearest neighbor (K-NN). Among these algorithms, the DT provided better results. In addition, the most important factors influencing the great customer experience were extracted with the help of the DT. Finally, very interesting results were observed in terms of the effect of the number of features on the performance of machine learning algorithms. [ABSTRACT FROM AUTHOR]

The convolutional neural networks (CNNs) have proven to be powerful classification tools in tasks that range from check reading to medical diagnosis, reaching close to human perception, and in some cases surpassing it. However, the problems to solve are becoming larger and more complex, which translates to larger CNNs, leading to longer training times that not even the adoption of Graphics Processing Units (GPUs) could keep up to. This problem is partially solved by using more processing units and distributed training methods that are offered by several frameworks dedicated to neural network training, such as Caffe, Torch, or TensorFlow. However, these techniques do not take full advantage of the possible parallelization offered by CNNs and the cooperative use of heterogeneous devices with different processing capabilities, clock speeds, memory size, among others. This paper presents a new method for the parallel training of CNNs where only the convolutional layer is distributed. The paper analyzes the influence of network size, bandwidth, batch size, number of devices, including their processing capabilities, and other parameters. Results show that this technique is capable of diminishing the training time without affecting the classification performance for both CPUs and GPUs. For the CIFAR-10 dataset, using a CNN with two convolutional layers, and and kernels, respectively, best speedups achieve using four CPUs and with three GPUs. Larger datasets will certainly require more than -% of processing time calculating convolutions, and speedups will tend to increase accordingly. [ABSTRACT FROM AUTHOR]

This paper presents an enrollment management model by applying artificial neural network (ANN). The aim of the research, which has been presented in this paper, is to show that ANNs are more successful in predicting than the classical statistical method - regression analysis (logistic regression). Both predictive models, no matter whether they are based on ANNs or logistic regression, offer satisfactory predictive results, and they can offer support in the decision-making process. However, the model based on neural networks shows certain advantages. ANNs demand understanding of functional connection between independent and dependent variables in order to evaluate the model. Also, they adapt easily to related independent variables, without the appearance of the problem of multicollinearity. In contrast to logistic regression, neural networks can recognize the appearance of nonlinearity and interactions in input data, and they can react on time. [ABSTRACT FROM AUTHOR]

Text Classification has received significant attention in recent years because of the proliferation of digital documents and is widely used in various applications such as filtering and recommendation. Consequently, many approaches, including those based on statistical theory, machine learning, and classifier performance improvement, have been proposed for improving text classification performance. Among these approaches, centroid-based classifier, multinomial naïve bayesian (MNB), support vector machines (SVM), convolutional neural network (CNN) are commonly used. In this paper, we introduce a cosine similarity-based methodology for improving performance. The methodology combines cosine similarity (between a test document and fixed categories) with conventional classifiers such as MNB, SVM, and CNN to improve the accuracy of the classifiers, and then we call the conventional classifiers with cosine similarity as enhanced classifiers. We applied the enhanced classifiers to famous datasets – 20NG, R8, R52, Cade12, and WebKB – and evaluated the performance of the enhanced classifiers in terms of the confusion matrix's accuracy; we obtained outstanding results in that the enhanced classifiers show significant increases in accuracy. Moreover, through experiments, we identified which of two considered knowledge representation techniques (word count and term frequency-inverse document frequency (TFIDF)) is more suitable in terms of classifier performance. [ABSTRACT FROM AUTHOR]

Video-surveillance has always been a vital tool to enforce safety in both public and private environments. Even though (smart) cameras are nowadays relatively widespread and cheap, such monitoring systems lack effectiveness in most scenarios. In addition, there is no guarantee about a human operator who monitors rare events in live video footages, forcing the use of such systems after unwanted events already took their undisturbed course, as a mere tool for investigations. Having an intelligent software to perform the task would allow to unlock the full potential of video-surveillance systems. To this end, in this paper we propose a solution based on a 3D Convolutional Neural Network that can effectively detect fights, aggressive motions and violence scenes in live video streams. Compared to state-of-the-art techniques, our method showed very promising performance on three challenging benchmark datasets: Hockey Fight, Crowd Violence and Movie Violence. [ABSTRACT FROM AUTHOR]

Neural tensor network (NTN) has been recently introduced to complete Resource Description Framework (RDF) knowledge bases, which has been the state-of-the-art in the field so far. An RDF knowledge base includes some facts from the real world shown as RDF "triples." In the previous methods, an objective function has been used for training this type of network, and the network parameters should have been set in a way to minimize the function. For this purpose, a classic nonlinear optimization method has been used. Since many replications are needed in this method to get the minimum amount of the function, in this paper, we suggest to combine meta-heuristic optimization methods to minimize the replications and increase the speed of training consequently. So, this problem will be improved using some meta-heuristic algorithms in this new approach to specify which algorithm will get the best results on NTN and its results will be compared with the results of the former methods finally. [ABSTRACT FROM AUTHOR]

Where, when and how much animals eat provide valuable insights into their ecology. In this paper, we present a comparative analysis between Support Vector Machine (SVM) and Input Delay Neural Network (IDNN) models to identify prey capture events from penguin accelerometry data. A pre-classified dataset of 3D time-series data from back-mounted accelerometers was used. We trained both the models to classify the penguins’ behavior at intervals as either ‘prey handling’ or ‘swimming’. The aim was to determine whether IDNN could achieve the same level of classification accuracy as SVM, but with reduced memory demands. This would enable the IDNN model to be embedded on the accelerometer micro-system itself, and hence reduce the magnitude of the output data to be uploaded. Based on the classification results, this paper provides an analysis of the two models from both an accuracy and applicability point of view. The experimental results show that both models achieve an equivalent accuracy of approx. 85% using the featured data, with a memory demand of 0.5 kB for IDNN and 0.7 Mb for SVM. The raw accelerometer data let us improve the generalizability of the models with a slightly lower accuracy to around 80%. This indicates that the IDNN model can embed on the accelerometer itself, reducing problems associated with raw time-series data retrieval and loss. [ABSTRACT FROM AUTHOR]

In machine learning driven surface inspection one often faces the issue that defects to be detected are difficult to make available for training, especially when pixel-wise labeling is required. Therefore, supervised approaches are not feasible in many cases. In this paper, this issue is circumvented by injecting synthetized defects into fault-free surface images. In this way, a fully convolutional neural network was trained for pixel-accurate defect detection on decorated plastic parts, reaching a pixel-wise PRC score of 78% compared to 8% that was reached by a state-of-the-art unsupervised anomaly detection method. In addition, it is demonstrated that a similarly good performance can be reached even when the network is trained on only five fault-free parts. [ABSTRACT FROM AUTHOR]

An automated system for early diagnosis of type 2-diabetes mellitus is proposed in this paper, by using the Extreme Learning Machine neural network for classification and the evolutionary genetic algorithms for feature extraction, to be employed on a real data set from Saudi Arabian patients. The dimension of the feature space is reduced by the genetic algorithms and only the effective features are selected. The data is then fed to an Extreme Learning Machine neural network for classification. Diabetes is a major health problem in both industrial and developing countries, and when it appears in pregnancies it may cause many complications, hence its early diagnosis is beneficial for both mother and fetus. Our hybrid algorithm, the GA-ELM algorithm, has produced an optimized diagnosis of type 2-diabetes patients and classified the data set with an accuracy of 97.5% and with six effective features, out of the original eight features given in the dataset. Moreover, comparisons of the GA-ELM method with other available methods were conducted and the results are promising. [ABSTRACT FROM AUTHOR]

Generally, in neural networks the sparseness is a suitable regularizer in a lot of applications. In this paper, sparse connectivity and sparse representation are used to enhance solutions to the problem of classification. Sequential feature selection is then leveraged to remove redundant features and select relevant ones. Sparseness-enforcing projection operator is used to discovering the most similar vector with a predefined sparseness degree for any input vector as well. As it will be argued, the mentioned operator is approximately differentiable at every point. From the facts it is clear that the sparseness enforcing projection would be appropriate for use as a transfer function in the proposed neural network and the network can be tuned using gradient based methods. Meanwhile, an intelligent method was used to build the architecture of the proposed neural network to achieve better performance. The MNIST dataset which consists of 70,000 handwritten digits was used to train and test the method and 99.18% accuracy was achieved by classifying this dataset. [ABSTRACT FROM AUTHOR]

There are known applications of convolutional neural networks to vision inspection of natural products. For many products it is sufficient to acquire and process a single image, but some might require imaging from two sides. Human experts performing quality inspection of malting barley typically only observe one side of each grain, but in doubtful cases look at both sides, intrinsically combining the information. In this paper, we make two contributions. We present a method for determining whether imaging objects from two sides yields performance benefits over single-sided imaging. Then we introduce a double-stream convolutional network for reasoning from two images simultaneously and analyze several methods of combining information from two streams. We find that when orientation of the object is unpredictable and the streams are not specialized to process a particular view, a fully shared architecture combining information on the prediction level yields best performance (98.7% accuracy on our dataset). [ABSTRACT FROM AUTHOR]

This paper highlights modeling affective temperature control in food small and medium-sized enterprises (SMEs). Modeling defined that workstation temperature set point could be controlled based on worker heart rate and workstation environment using Artificial Neural Network (ANN). The research objectives were: 1) to propose modeling affective temperature control in food SMEs based on heart rate and workstation environment; and 2) to develop an ANN model for predicting workstation temperature set point. Training and validation data were collected from six food SMEs in Yogyakarta Special Region, Indonesia. The data of temperature set points were verified using a simulated confined room. The inputs of the ANN model were worker heart rate, workstation temperature, relative humidity distribution and light intensity. The output was temperature set point. Research results concluded satisfactory performance of ANN. The model could be used to provide environmental ergonomics in food SMEs. [ABSTRACT FROM AUTHOR]

Passenger flow forecasting is an essential part of transportation systems. Neural networks in the transportation field have been applied to passenger demand prediction. In this paper, we developed two hybrid methods, known as parlimentary optimization algorithm-artificial neural network (POA-ANN), and intelligent water drops algorithm-ANN (IWD algorithm-ANN). In addition, we applied the proposed algorithms to illustrate the effect of precise prediction for passenger queues. We mainly focus on predicting passenger demand by comparing the genetic algorithm-ANN (GA-ANN) with POA-ANN and IWD-ANN. The results of prediction methods suggest that both POA-ANN and IWD-ANN provide a better forecasting performance, which is obtained via mean square error (MSE), than GA-ANN in the field of passenger flow prediction. This study illustrates that the newly adopted algorithms exhibit good performance for passenger prediction. [ABSTRACT FROM AUTHOR]