Different Sizes Of Training Sets Research Articles

Fault Detection and Failure Rate Analysis of New Energy Vehicles Based on Decision Tree Algorithm

Abstract New energy vehicles are vital in promoting environmental protection and technological innovation. Fault detection still faces challenges during its operation, and efficient and accurate methods for fault diagnosis are urgently needed. This paper proposes a fault detection and analysis model based on a decision tree algorithm for the fault detection problem of new energy vehicles. The dataset applicable to the model is prepared by preprocessing in-vehicle network data, including data cleaning, integration, and other steps. Fault prediction can be realized after using C4.5 algorithms to construct a decision tree. With a precision of 82.26% on the test set, this model is highly accurate in fault detection, which is 1.23 percentage points higher than the traditional decision tree algorithm. The model’s effectiveness and efficiency in handling large-scale data were demonstrated by its training and testing on training sets of different sizes. Using the traditional algorithm, a training set of 80,000 data was used to reduce the model’s running time from 274,432 seconds to 249,269 seconds. This study provides a practical methodology for fault diagnosis of new energy vehicles, improving fault detection accuracy while optimizing computational efficiency. Real-time monitoring and timely maintenance of new energy vehicles require this.

Higher-order link prediction via local information.

Link prediction has been widely studied as an important research direction. Higher-order link prediction has gained, in particular, significant attention since higher-order networks provide a more accurate description of real-world complex systems. However, higher-order networks contain more complex information than traditional pairwise networks, making the prediction of higher-order links a formidable challenging task. Recently, researchers have discovered that local features have advantages over long-range features in higher-order link prediction. Therefore, it is necessary to develop more efficient and concise higher-order link prediction algorithms based on local features. In this paper, we proposed two similarity metrics via local information, simplicial decomposition weight and closed ratio weight, to predict possible future higher-order interactions (simplices) in simplicial networks. These two algorithms capture local higher-order information at two aspects: simplex decomposition and cliques' state (closed or open). We tested their performance in eight empirical simplicial networks, and the results show that our proposed metrics outperform other benchmarks in predicting third-order and fourth-order interactions (simplices) in most cases. In addition, we explore the robustness of the proposed algorithms, and the results suggest that the performance of these novel algorithms is advanced under different sizes of training sets.

Simultaneous measurement of chemical oxygen demand and turbidity in water based on broad optical spectra using backpropagation neural network

How to overcome the measurement interference between chemical oxygen demand (COD) and turbidity in absorption spectra of water has attracted lots of attention recently. Previous studies focused on eliminating the influence of turbidity, which could not measure COD and turbidity simultaneously. In this paper, we proposed a new method based on ultraviolet-visible-near infrared (UV-Vis-NIR) absorption measurements using a backpropagation neural network (BPNN). The BPNN model was established within the Tensorflow framework. And the absorbance of the wavelength ranges of 240–1040 nm and the actual values of COD and turbidity in the training set were defined as input and output, respectively. Then the BPNN model was trained with training sets of different sizes. And the original absorbance in the test set without any compensation was used to test the performance of the trained BPNN, which indicated that the larger the sample size, the better the adjustment effect of the model. The optimal training sample size was 2400. The correlation coefficient of the model is greater than 0.9998, and the RMSE between the predicted value and the actual value of COD and turbidity is 0.19 mg/L and 0.34 NTU, respectively. Our results show that the method based on BPNN not only can eliminate the influence of turbidity on the accuracy of COD but also realize the accurate measurement of turbidity itself. Furthermore, this method provides a potential strategy for the simultaneous and rapid measurement of other water quality parameters by using the absorption spectrum.

Instance Selection-Based Surrogate-Assisted Genetic Programming for Feature Learning in Image Classification.

Genetic programming (GP) has been applied to feature learning for image classification and achieved promising results. However, many GP-based feature learning algorithms are computationally expensive due to a large number of expensive fitness evaluations, especially when using a large number of training instances/images. Instance selection aims to select a small subset of training instances, which can reduce the computational cost. Surrogate-assisted evolutionary algorithms often replace expensive fitness evaluations by building surrogate models. This article proposes an instance selection-based surrogate-assisted GP for fast feature learning in image classification. The instance selection method selects multiple small subsets of images from the original training set to form surrogate training sets of different sizes. The proposed approach gradually uses these surrogate training sets to reduce the overall computational cost using a static or dynamic strategy. At each generation, the proposed approach evaluates the entire population on the small surrogate training sets and only evaluates ten current best individuals on the entire training set. The features learned by the proposed approach are fed into linear support vector machines for classification. Extensive experiments show that the proposed approach can not only significantly reduce the computational cost but also improve the generalisation performance over the baseline method, which uses the entire training set for fitness evaluations, on 11 different image datasets. The comparisons with other state-of-the-art GP and non-GP methods further demonstrate the effectiveness of the proposed approach. Further analysis shows that using multiple surrogate training sets in the proposed approach achieves better performance than using a single surrogate training set and using a random instance selection method.

An Online Electricity Market Price Forecasting Method Via Random Forest

Electricity price forecasting (EPF) is essential to the bidding strategy formulation and market operation. Since EPF is important in the electricity market, lots of forecasting approaches are proposed. However, the new scene caused by the volatility of renewable power generation and other volatility factors has made previous methods inaccurate and inapplicable. To address this problem, we propose an online self-adaptive forecasting method based on random forest, which is different from the traditional batch learning. Our approach takes possible fluctuations of the market into consideration, and adapts to them by maintaining training sets of different sizes. A case study using actual electricity market data has shown that our proposed approach obtains higher accuracy than ordinary approaches, as well as sheds light on possible concept drift in the market.

A Random CNN Sees Objects: One Inductive Bias of CNN and Its Applications

This paper starts by revealing a surprising finding: without any learning, a randomly initialized CNN can localize objects surprisingly well. That is, a CNN has an inductive bias to naturally focus on objects, named as Tobias ("The object is at sight") in this paper. This empirical inductive bias is further analyzed and successfully applied to self-supervised learning (SSL). A CNN is encouraged to learn representations that focus on the foreground object, by transforming every image into various versions with different backgrounds, where the foreground and background separation is guided by Tobias. Experimental results show that the proposed Tobias significantly improves downstream tasks, especially for object detection. This paper also shows that Tobias has consistent improvements on training sets of different sizes, and is more resilient to changes in image augmentations.

Two-step discriminant analysis based multi-view polarimetric SAR image classification with high confidence

Polarimetric synthetic aperture radar (PolSAR) image classification is a hot topic in remote sensing field. Although recently many deep learning methods such as convolutional based networks have provided great success in PolSAR image classification, but they need a high volume of labeled samples, which are not usually available in practice, or they cause a high computational burden for implementation. In this work, instead of spending cost for network training, the inherent nature of PolSAR image is used for generation of convolutional kernels for extraction of deep and robust features. Moreover, extraction of diverse scattering characteristics contained in the coherency matrix of PolSAR and fusion of their output classification results with a high confidence have high impact in providing a reliable classification map. The introduced method called discriminative features based high confidence classification (DFC) utilizes several approaches to deal with difficulties of PolSAR image classification. It uses a multi-view analysis to generate diverse classification maps with different information. It extracts deep polarimetric-spatial features, consistent and robust with respect to the original PolSAR data, by applying several pre-determined convolutional filters selected from the important regions of image. Convolutional kernels are fixed without requirement to be learned. The important regions are determined with selecting the key points of image. In addition, a two-step discriminant analysis method is proposed to reduce dimensionality and result in a feature space with minimum overlapping and maximum class separability. Eventually, a high confidence decision fusion is implemented to find the final classification map. Impact of multi-view analysis, selection of important regions as fixed convolutional kernels, two-step discriminant analysis and high confidence decision fusion are individually assessed on three real PolSAR images in different sizes of training sets. For example, the proposed method achieves 96.40% and 98.72% overall classification accuracy by using 10 and 100 training samples per class, respectively in L-band Flevoland image acquired by AIRSAR. Generally, the experiments show high efficiency of DFC compared to several state-of-the-art methods especially for small sample size situations.

A Transformer Based Encodings for Detection of Semantically Equivalent Questions in cQA

Abstract The probability of redundancy in questions has significantly increased due to the increasing influx of users on different cQA forums such as Quora, Stack overflow, etc. Because of this redundancy, the responses are scattered through various variations of the same question that results in unsatisfactory search results to a specific question. To address this issue, this work proposes the model for discovering the semantic similarity among the cQA questions. We followed two approaches (i) Feature-based: the question embedding is created using four forms of word embeddings and an ensemble of all four. Then Siamese LSTM (sLSTM) is used to find the semantic similarity among the questions. (ii) Fine-tuning: we fine-tuned BERT model on STS and SNLI data, which employs Siamese network architectures to generate semantically meaningful sentence embeddings. Then sBERT is used to assess the similarity between the questions. Experiments were carried out on Quora (QQP) and Stack Exchange cQA dataset with training sets of different sizes and word vectors of different dimensionalities. The model shows significant improvement over the state-of-the-artwork on sentence similarity tasks.

Classification algorithms of error-related potentials in brain-computer interface

Error self-detection based on error-related potentials (ErrP) is promising to improve the practicability of brain-computer interface systems. But the single trial recognition of ErrP is still a challenge that hinters the development of this technology. To assess the performance of different algorithms on decoding ErrP, this paper test four kinds of linear discriminant analysis algorithms, two kinds of support vector machines, logistic regression, and discriminative canonical pattern matching (DCPM) on two open accessed datasets. All algorithms were evaluated by their classification accuracies and their generalization ability on different sizes of training sets. The study results show that DCPM has the best performance. This study shows a comprehensive comparison of different algorithms on ErrP classification, which could give guidance for the selection of ErrP algorithm.

Big Data Aspect-Based Opinion Mining Using the SLDA and HME-LDA Models

In order to make better use of massive network comment data for decision-making support of customers and merchants in the big data era, this paper proposes two unsupervised optimized LDA (Latent Dirichlet Allocation) models, namely, SLDA (SentiWordNet WordNet-Latent Dirichlet Allocation) and HME-LDA (Hierarchical Clustering MaxEnt-Latent Dirichlet Allocation), for aspect-based opinion mining. One scheme of each of two optimized models, which both use seed words as topic words and construct the inverted index, is designed to enhance the readability of experiment results. Meanwhile, based on the LDA topic model, we introduce new indicator variables to refine the classification of topics and try to classify the opinion target words and the sentiment opinion words by two different schemes. For better classification effect, the similarity between words and seed words is calculated in two ways to offset the fixed parameters in the standard LDA. In addition, based on the SemEval2016ABSA data set and the Yelp data set, we design comparative experiments with training sets of different sizes and different seed words, which prove that the SLDA and the HME-LDA have better performance on the accuracy, recall value, and harmonic value with unannotated training sets.

Microseismic records classification using capsule network with limited training samples in underground mining

The identification of suspicious microseismic events is the first crucial step in microseismic data processing. Existing automatic classification methods are based on the training of a large data set, which is challenging to apply in mines without a long-term manual data processing. In this paper, we present a method to automatically classify microseismic records with limited samples in underground mines based on capsule networks (CapsNet). We divide each microseismic record into 33 frames, then extract 21 commonly used features in time and frequency from each frame. Consequently, a 21 × 33 feature matrix is utilized as the input of CapsNet. On this basis, we use different sizes of training sets to train the classification models separately. The trained model is tested using the same test set containing 3,200 microseismic records and compared to convolutional neural networks (CNN) and traditional machine learning methods. Results show that the accuracy of our proposed method is 99.2% with limited training samples. It is superior to CNN and traditional machine learning methods in terms of Accuracy, Precision, Recall, F1-Measure, and reliability.

Choosing an appropriate training set size when using existing data to train neural networks for land cover segmentation

ABSTRACT Land cover data is an inventory of objects on the Earth’s surface, which is often derived from remotely sensed imagery. Deep Convolutional Neural Network (DCNN) is a competitive method in image semantic segmentation. Some scholars argue that the inadequacy of training set is an obstacle when applying DCNNs in remote sensing image segmentation. While existing land cover data can be converted to large training sets, the size of training data set needs to be carefully considered. In this paper, we used different portions of a high-resolution land cover map to produce different sizes of training sets to train DCNNs (SegNet and U-Net) and then quantitatively evaluated the impact of training set size on the performance of the trained DCNN. We also introduced a new metric, Edge-ratio, to assess the performance of DCNN in maintaining the boundary of land cover objects. Based on the experiments, we document the relationship between the segmentation accuracy and the size of the training set, as well as the nonstationary accuracies among different land cover types. The findings of this paper can be used to effectively tailor the existing land cover data to training sets, and thus accelerate the assessment and employment of deep learning techniques for high-resolution land cover map extraction.

QSPR modelling of the soil sorption coefficient from training sets of different sizes

ABSTRACTQuantitative structure–property relationship (QSPR) modelling has been used in many scientific fields. This approach has been extensively applied in environmental research to predict physicochemical properties of compounds with potential environmental impact. The soil sorption coefficient is an important parameter for the evaluation of environmental risks, and it helps to determine the final fate of substances in the environment. In the last few years, different QSPR models have been developed for the determination of the sorption coefficient. In this study, several QSPR models were generated and evaluated for the prediction of log Koc from the relationship with log P. These models were obtained from an extensive and diverse training set (n = 639) and from subsets of this initial set (i.e. halves, fourths and eighths). The aim of this study was to investigate whether the size of the training set affects the statistical quality of the obtained models. Furthermore, statistical equivalence was verified between the models obtained from smaller sets and the model obtained from the total training set. The results confirmed the equivalence between the models, thus indicating the possibility of using smaller training sets without compromising the statistical quality and predictive capability, as long as most chemical classes in the test set are represented in the training set.

Automatically configuring the reference point method for automated multi-objective treatment planning

Automated treatment planning algorithms have demonstrated capability in generating consistent and high-quality treatment plans. Their configuration (i.e. determining the algorithm’s parameters), however, can be a labour-intensive and time-consuming trial-and-error procedure.Previously, we introduced the reference point method (RPM) for fast automated multi-objective treatment planning. The RPM generates a single Pareto optimal plan for each patient. When the RPM is configured appropriately, this plan has clinically favourable trade-offs between all plan objectives. This paper proposes a new procedure to automatically generate a single configuration of the RPM per tumour site. The procedure was tested for prostate cancer.Planning CT scans of 287 previously treated patients were included in a database, together with corresponding Pareto optimal plans generated using our clinically applied two-phase -constraint method (part of Erasmus-iCycle) for automated multi-objective treatment planning. The procedure developed acquires plan characteristics observed in a training set. Based on these, an RPM configuration is automatically generated according to user preferences which specify acceptable differences between training set plans and corresponding RPM generated plans. For example, compared to the training set plans, the RPM generated plans need to have similar PTV coverage, and preferably reduced high rectum dose while slight deteriorations in other objectives are allowed. Training sets of different sizes were tested, and the quality of the resulting RPM configurations was evaluated on the test set (subset of the database not used for training).Using the new procedure, an RPM configuration was generated for each training set. The quality of RPM generated plans was similar or slightly better than that of the corresponding test set plans. The proposed automated configuration procedure greatly reduces the manual configuration workload, and thereby improves the efficiency and effectiveness of an automated clinical treatment planning workflow.

Intelligent simultaneous fault diagnosis for solid oxide fuel cell system based on deep learning

Fault diagnosis technology is a vital tool for ensuring the stability and durability of solid oxide fuel cell systems. Simultaneous faults are common problems in modern industrial systems. Many fault diagnosis methods have been successfully designed for solid oxide fuel cell systems, but they only address independent faults, and only a few researchers have studied simultaneous fault diagnosis. The design of a simultaneous fault diagnosis method for solid oxide fuel cell systems remains a huge challenge. This study introduces a deep learning technology into the simultaneous fault diagnosis for the solid oxide fuel cell system and proposes a novel simultaneous fault diagnosis method on the basis of a deep learning network called stacked sparse autoencoder. The proposed method can automatically capture the essential features from the original system variables, thereby consuming minimal time on heavily hand-crafted features. Moreover, massive unlabeled samples are fully utilized through the proposed method. Experimental results show that the proposed method can diagnose simultaneous faults with high accuracy requiring only a few independent fault samples and a minimal number of simultaneous fault samples. Comparisons between traditional machine learning methods and experimental results on training sets of different sizes verify the superiority of the proposed method. Deep learning provides an effective and promising approach for simultaneous fault diagnosis in the field of fuel cells.

Utilization of alternating conditional expectation method for pin power reconstruction in CANDU reactor physics analyses

This paper presents a new methodology using the Alternating Conditional Expectation (ACE) algorithm for reconstructing fuel pin powers in the Canada Deuterium Uranium (CANDU) reactor. Variations in assembly powers in neighboring fuel bundles are used as predictor variables to determine the response, e.g. power in each of the fuel pin. When a sufficient number of cases have been evaluated, the relation between the predictor variables and the response can be treated as a multi-variable regression problem. The ACE algorithm which is a non-parametric approach (i.e., no predetermined functional relationship between the predictors and the response is assumed) is then utilized for estimating optimal transformations of the predictors and the response. The geometry used is a two-dimensional, 3-by-3 fuel bundles configuration. Three hundred different configurations, which represent various power distributions, were evaluated to develop these optimal transformations. The optimal transformation is determined for each of 37 pins in a CANDU natural uranium fuel bundle. The quality of these transformations for predicting fuel pin power has been verified against an independent set of 50 two-dimensional, 3-by-3 fuel bundle configurations. The results indicate that the proposed pin power reconstruction algorithm is sufficiently accurate, giving an RMS (root-mean-squared) error of around 0.59% for all pins. The RMS error in predicting the maximum pin power is 0.60%. The development of the proposed methodology continued with the examination of its robustness. The impact of employing different training sets on the overall relative error in predicting the pin power is examined. In addition, the impact of employing different sizes of training sets is also investigated. For each training set size, a bias in predicting the maximum pin power is determined statistically based on the numerical results observed during the study. The results indicate that the methodology is likely to under-predict the maximum pin power; to circumvent this issue a multiplicative bias of 1.0118 needs to be applied, based on evaluating results from various sizes of the training sets.

Link prediction in microblog retweet network based on maximum entropy model

Microblog is a social media platform, based on the follower-followee relationship, that enables users to share real-time information, by which the information propagation is characterized as rapid, explosive, and immediate. The research on the information propagation and retweet prediction is very important for public sentiment analysis and product promotion. A majority of existing works adopt several traditional prediction methods to predict the future information retweet based on the features extracted from existing retweet behaviors, which are hard to reconcile accuracy, complexity, robustness and feature extensiveness. To overcome the above mentioned shortcomings in existing works, we propose in this paper a link prediction algorithm based on maximum entropy model to predict retweet behavior on microblog. In our proposed approach, firstly we abstract the retweet prediction problem to a link prediction problem. Then we analyze the retweet behaviors on microblog and determine the factors influencing the retweet behavior. We extract the features from the retweet behaviors based on these factors in the next step. Now based on these features, the retweet behavior could be predicted by the proposed approach. However, information redundancy and other issues may exist among these features. These issues will cause an increase in computational complexity or a decrease in computational accuracy. To solve the above problems, we selecte the features dominating the retweet behavior with feature selection methods such as Information Gain, IG-CHI. The proposed model requires no further independent assumption in features or intrinsic constraints, and omits the processing in relation to features, which is usually the prerequisite of other prediction methods. We take the Sina Weibo retweet records in a time span from 2009 to 2012 as an example to test the effectiveness and efficiency of our link prediction algorithm. Results show that: 1) the proposed algorithm has incomparable advantages in running time; 2) as for the predicted result, the proposed algorithm is better than other algorithms in performance evaluations; 3) the proposed algorithm runs stably for different sizes of training sets and feature sets; 4) the accuracy of the predicted results remains stable based on the selected features. The proposed approach avoids the independent restriction among features and shows better accuracy than other similar methods, thus it has reference values for resolving other prediction problems in complex networks.

Automated systems to identify relevant documents in product risk management

BackgroundProduct risk management involves critical assessment of the risks and benefits of health products circulating in the market. One of the important sources of safety information is the primary literature, especially for newer products which regulatory authorities have relatively little experience with. Although the primary literature provides vast and diverse information, only a small proportion of which is useful for product risk assessment work. Hence, the aim of this study is to explore the possibility of using text mining to automate the identification of useful articles, which will reduce the time taken for literature search and hence improving work efficiency. In this study, term-frequency inverse document-frequency values were computed for predictors extracted from the titles and abstracts of articles related to three tumour necrosis factors-alpha blockers. A general automated system was developed using only general predictors and was tested for its generalizability using articles related to four other drug classes. Several specific automated systems were developed using both general and specific predictors and training sets of different sizes in order to determine the minimum number of articles required for developing such systems.ResultsThe general automated system had an area under the curve value of 0.731 and was able to rank 34.6% and 46.2% of the total number of 'useful' articles among the first 10% and 20% of the articles presented to the evaluators when tested on the generalizability set. However, its use may be limited by the subjective definition of useful articles. For the specific automated system, it was found that only 20 articles were required to develop a specific automated system with a prediction performance (AUC 0.748) that was better than that of general automated system.ConclusionsSpecific automated systems can be developed rapidly and avoid problems caused by subjective definition of useful articles. Thus the efficiency of product risk management can be improved with the use of specific automated systems.

Optimization of Radial Basis Function neural network employed for prediction of surface roughness in hard turning process using Taguchi’s orthogonal arrays

This work presents a study on the applicability of radial base function (RBF) neural networks for prediction of Roughness Average (Ra) in the turning process of SAE 52100 hardened steel, with the use of Taguchi’s orthogonal arrays as a tool to design parameters of the network. Experiments were conducted with training sets of different sizes to make possible to compare the performance of the best network obtained from each experiment. The following design factors were considered: (i) number of radial units, (ii) algorithm for selection of radial centers and (iii) algorithm for selection of the spread factor of the radial function. Artificial neural networks (ANN) models obtained proved capable to predict surface roughness in accurate, precise and affordable way. Results pointed significant factors for network design have significant influence on network performance for the task proposed. The work concludes that the design of experiments (DOE) methodology constitutes a better approach to the design of RBF networks for roughness prediction than the most common trial and error approach.

Hybrid artificial neural network system for short-term load forecasting

This paper presents a novel hybrid method for Short-Term Load Forecasting (STLF). The system comprises of two Artificial Neural Networks (ANN), assembled in a hierarchical order. The first ANN is a Multilayer Perceptron (MLP) which functions as integrated load predictor (ILP) for the forecasting day. The output of the ILP is then fed to another, more complex MLP, which acts as an hourly load predictor (HLP) for a forecasting day. By using a separate ANN that predicts the integral of the load (ILP), additional information is presented to the actual forecasting ANN (HLP), while keeping its input space relatively small. This property enables online training and adaptation, as new data become available, because of the short training time. Different sizes of training sets have been tested, and the optimum of 30 day sliding time-window has been determined. The system has been verified on recorded data from Serbian electrical utility company. The results demonstrate better efficiency of the proposed method in comparison to non-hybrid methods because it produces better forecasts and yields smaller mean average percentage error (MAPE).