Sparse Datasets Research Articles

An Analysis of ML-Based Outlier Detection from Mobile Phone Trajectories

This paper provides an analysis of two machine learning algorithms, density-based spatial clustering of applications with noise (DBSCAN) and the local outlier factor (LOF), applied in the detection of outliers in the context of a continuous framework for the detection of points of interest (PoI). This framework has as input mobile trajectories of users that are continuously fed to the framework in close to real time. Such frameworks are today still in their infancy and highly required in large-scale sensing deployments, e.g., Smart City planning deployments, where individual anonymous trajectories of mobile users can be useful to better develop urban planning. The paper’s contributions are twofold. Firstly, the paper provides the functional design for the overall PoI detection framework. Secondly, the paper analyses the performance of DBSCAN and LOF for outlier detection considering two different datasets, a dense and large dataset with over 170 mobile phone-based trajectories and a smaller and sparser dataset, involving 3 users and 36 trajectories. Results achieved show that LOF exhibits the best performance across the different datasets, thus showing better suitability for outlier detection in the context of frameworks that perform PoI detection in close to real time.

Integrating Collaborative Filtering Technique Using Rating Approach to Ascertain Similarity Between the Users

The recommender system handles the plethora of data by filtering the most crucial information based on the dataset provided by a user and other criterion that are taken into account.(i.e., user's choice and interest). It determines whether a user and an item are compatible and then assumes that they are similar in order to make recommendations. Recommendation system uses Singular value decomposition method as collaborative filtering technique. The objective of this research paper is to propose the recommendation system that has an ability to recommend products to users based on ratings. We collect essential information like ratings given by the users from e-commerce that are required for recommendation, Initially the dataset that are gathered are sparse dataset, cosine similarity is used to find the similarity between the users. Subsequently, we collect non-sparse data and use Euclidian distance and Manhattan distance method to measure the distance between users and the graph is plotted, this ensures the similar liking and preferences between them. This method of making recommendations are more reliable and attainable.

Advantages of a Statistical Estimation Approach for Clock Frequency Estimation of Heterogeneous and Irregular CGRAs

Estimating the maximum clock frequency of homogeneous Coarse Grained Reconfigurable Arrays/Architectures (CGRAs) with an arbitrary number of Processing Elements (PE) is difficult. Clock frequency estimation of highly heterogeneous CGRAs takes additional factors into account, thus is even more difficult. Main challenges are the heterogeneous set of operators for each Processing Element (PE) and the irregular interconnect (connecting a CGRA’s PEs). Multiple estimation approaches could be reasonable. We propose an optimized statistical estimator, which is based on our prior work. We demonstrate its superiority to state-of-the-art neural networks in terms of accuracy and robustness, especially in situations with a sparse set of training data.

Data driven modeling of interfacial traction–separation relations using a thermodynamically consistent neural network

For multilayer structures, interfacial failure is one of the most important elements related to device reliability. For cohesive zone modeling, traction–separation relations represent the adhesive interactions across interfaces. However, existing theoretical models do not currently capture traction–separation relations that have been extracted using direct methods, particularly under mixed-mode conditions. Given the complexity of the problem, models derived from the neural network approach are attractive. Although they can be trained to model data along the loading paths taken in a particular set of mixed-mode fracture experiments, they may fail to obey physical laws for paths not covered by the training data sets. In this paper, a thermodynamically consistent neural network (TCNN) approach is established to model the constitutive behavior of interfaces when faced with sparse training data sets. Accordingly, four conditions are examined and implemented here: (i), positive energy dissipation; (ii), maximum energy 10 dissipation; (iii), conservation of the J-integral; (iv), dependence of interfacial toughness on the fracture mode-mix. These conditions are treated as constraints and are implemented as such in the loss function. The feasibility of this approach is demonstrated by comparing the modeling results with a range of physical constraints and several different input data sets. Moreover, a Bayesian optimization algorithm is then adopted to optimize the weight factors associated with each of the constraints in order to overcome convergence issues that can arise when multiple constraints are present. The resultant numerical implementation of the ideas presented here produced well-behaved, mixed-mode traction–separation​ surfaces that maintained the fidelity of the experimental data that was provided as input. Some guidance is also provided on the desirable features of input data sets. The proposed approach heralds a new autonomous, point-to-point constitutive modeling concept for interface mechanics.

Multi-scale Pix2Pix network for high-fidelity prediction of adiabatic cooling effectiveness in turbine cascade

Film cooling is one of the effective cooling methods to ensure the longevity of high thermal load turbines. Due to multiple corresponding design parameters, it is difficult to seek rapid evolution of overall film cooling performance of new design. Currently, some achievements were obtained in plane cooling by implementing deep learning models which have strong nonlinear mapping capability in high-dimensional datasets. To further expand deep learning and achieve high-fidelity prediction on 3D complex cooling configurations, our research introduces deep learning network into the linear cascade of air-cooling turbines. Furthermore, in this work, the Multi-scale Pixel to Pixel (MSPix2Pix) network is proposed to realize the reconstruction of high-resolution adiabatic cooling effectiveness on turbine cascade among sparse dataset, in which high-resolution non-parametric concept and introduce multiple generators and discriminators are utilized. The average structural similarity (SSIM) reached 0.9892 between the predicted images and the CFD results in the test set. The results of the verification experiment show the applicability of MSPix2Pix network for rapid and accurate evaluation of cooling effectiveness on complex three-dimensional geometry with a certain generalization, which provides a certain support for high fidelity prediction of cooling configuration and aero-thermal coupling design in gas-cooling turbine.

Social Media Based Deep Auto-Encoder Model for Clinical Recommendation

One of the most actively studied topics in modern medicine is the use of deep learning and patient clinical data to make medication and ADR recommendations. However, the clinical community still has some work to do in order to build a model that hybridises the recommendation system. As a social media learning based deep auto-encoder model for clinical recommendation, this research proposes a hybrid model that combines deep self-decoder with Top n similar co-patient information to produce a joint optimisation function (SAeCR). Implicit clinical information can be extracted using the network representation learning technique. Three experiments were conducted on two real-world social network data sets to assess the efficacy of the SAeCR model. As demonstrated by the experiments, the suggested model outperforms the other classification method on a larger and sparser data set. In addition, social network data can help doctors determine the nature of a patient's relationship with a co-patient. The SAeCR model is more effective since it incorporates insights from network representation learning and social theory.

Correlations between EEG and intestinal electrical stimulation.

Many diseases affect the autonomous nervous system and the central nervous system simultaneously, for example Parkinson's disease or irritable bowel syndrome. To study neurophysiologic interactions between the intestinal electrical activity and the electroencephalography (EEG) pattern of the brain, we combined intestinal electrical stimulation (IES) and non-invasive telemetric full-band DC EEG recordings in an acute pig-model. Intestinal motility was monitored with accelerometers. Brain activity was analyzed with regard to network driven phenomena like phase amplitude coupling (PAC) within two time-windows: 1 min after IES (early response) and 3 min after stimulation (late response). Here we present the results for two stimulation sites (small intestine, colon) and two parietal scalp-EEG channels (right and left somatosensory cortex region). Electrical stimulation consisted of a 30 or 130 Hz pulse. In summary, the PAC modulation index at a parietal EEG recording position is decreased after IES. This effect is in line with an inhibitory effect of our IES protocol regarding peristalsis. The surprisingly strong effects of IES on network driven EEG patterns may be translated into new therapeutic techniques and/or diagnostic tools in the future. Furthermore, analytic tools, operating on sparse datasets, may be ideally suited for the integration in implantable intestinal pacemakers as feedback system.

Applying neural networks-based modelling to the prediction of mineralization: A case-study using the Western Australian Geochemistry (WACHEM) database

Geochemical data collected and stored by government and private sectors are becoming increasingly complex and too large for manual interpretation. The use of automatic methods for identification of potentially spurious data and estimation of missing values in very sparse geochemical datasets can significantly improve our understanding of geological systems. Deep neural networks have recently achieved remarkable success in a wide range of applied problems. These methods do not require manual feature engineering and significantly outperform traditional machine learning algorithms when applied to large datasets. We present a deep learning based method for estimation of unknown sample analytes in geochemical data. This approach is entirely data-driven and, once the network is trained, delivers the results in real time by predicting the distribution of an unknown analyte in a single step. A case study on the Western Australian Geochemistry (WACHEM) database demonstrates the efficiency of the method. Base metals as well as many other metals show good predictive capability (average symmetric mean absolute percentage errors range from 20% to 26.1%). Silver, platinum and especially gold are found to be more difficult to predict. The estimation of rock types from geochemical data allows for validation of existing datasets and prediction of rock types directly from geochemical data when there is no other existing information. The results of this study can benefit mineral explorers by indicating exploration targets and highlighting gaps in existing data.

A Unified Framework for Automatic Distributed Active Learning.

We propose a novel unified frameork for automated distributed active learning (AutoDAL) to address multiple challenging problems in active learning such as limited labeled data, imbalanced datasets, automatic hyperparameter selection as well as scalability to big data. First, automated graph-based semi-supervised learning is conducted by aggregating the proposed cost functions from different compute nodes and jointly optimizing hyperparameters in both the classification and query selection stages. For dense datasets, clustering-based uncertainty sampling with maximum entropy (CME) loss is applied in the optimization. For sparse and imbalanced datasets, shrinkage optimized KL-divergence regularization and local selection based active learning (SOAR) loss are further naturally adapted in AutoDAL. The optimization is efficiently resolved by iteratively executing a genetic algorithm (GA) refined with a local generating set search (GSS) and solving an integer linear programming (ILP) problem. Moreover, we propose an efficient distributed active learning algorithm which is scalable for big data. The proposed AutoDAL algorithm is applied to multiple benchmark datasets and two real-world datasets including an electrocardiogram (ECG) dataset and a credit fraud detection dataset for classification. We demonstrate that the proposed AutoDAL algorithm is capable of achieving significantly better performance compared to several state-of-the-art AutoML approaches and active learning algorithms.

Innovation, scaling-up, and local development in peripheral regions: do establishments scale-up locally?

Abstract A growing number of researchers suggest that there is no necessary connection between local firm-level innovation and local development. There are two connected arguments: first, many analysts suggest local innovation should be understood as a social and institutional process: from this perspective, just focusing on firms is too narrow. Second, regional economists view firms – especially innovative ones – as geographically mobile, tending to move away from regions which don’t offer them adequate support or resources. In this paper, we consider this second approach, exploring the degree to which innovators in peripheral regions purchase local services and scale-up their operations locally – two mechanisms that should connect firm-level innovation to local development. We find that peripheral innovators are more likely to use non-local services; their external expansion is also more likely to occur non-locally. If these results, obtained from cross-sectional analysis of a sparse data-set, are indicative of persistent processes, then they shed light on why firm-level innovation is disconnected from regional development in peripheral regions.

On training non-uniform fuzzy partitions for function approximation using differential evolution: A study on fuzzy transform and fuzzy projection

This paper focuses on the use of differential evolution to improve the approximation properties of function approximation models based on fuzzy partitions. Two cases are considered: Fuzzy transform and Fuzzy projection, and the design of hybrid evolutionary fuzzy systems, is studied. Even though function approximation techniques based on fuzzy partitions have been well studied, few papers consider the problem of centroid selection of the basis functions. Thus, in most cases uniform fuzzy partitions are considered. By using an evolutionary algorithm a systematic approach on the selection of the partition, is provided. The optimisation problem involves the determination of the model parameters, which in our case are the fuzzy partition’s membership functions’ locations. The proposed method is tested on the scattered data approximation problem, in a regression sense, that is given a set of sparse data the latent function is approximated. Numerical studies on one and two-dimensional test functions demonstrate that the evolutionary algorithm based fuzzy projection displays high performance in terms of approximation error. Moreover, the proposed approach shows high approximation capabilities with a small number of basis functions. Comparison results, with uniform fuzzy partition models, neural networks and support vector machines, are provided.

Fast virtual view synthesis for an 8K 3D light-field display based on cutoff-NeRF and 3D voxel rendering.

Three-dimensional (3D) light-field displays can provide an immersive visual experience, which has attracted significant attention. However, the generating of high-quality 3D light-field content in the real world is still a challenge because it is difficult to capture dense high-resolution viewpoints of the real world with the camera array. Novel view synthesis based on CNN can generate dense high-resolution viewpoints from sparse inputs but suffer from high-computational resource consumption, low rendering speed, and limited camera baseline. Here, a two-stage virtual view synthesis method based on cutoff-NeRF and 3D voxel rendering is presented, which can fast synthesize dense novel views with smooth parallax and 3D images with a resolution of 7680 × 4320 for the 3D light-field display. In the first stage, an image-based cutoff-NeRF is proposed to implicitly represent the distribution of scene content and improve the quality of the virtual view. In the second stage, a 3D voxel-based image rendering and coding algorithm is presented, which quantify the scene content distribution learned by cutoff-NeRF to render high-resolution virtual views fast and output high-resolution 3D images. Among them, a coarse-to-fine 3D voxel rendering method is proposed to improve the accuracy of voxel representation effectively. Furthermore, a 3D voxel-based off-axis pixel encoding method is proposed to speed up 3D image generation. Finally, a sparse views dataset is built by ourselves to analyze the effectiveness of the proposed method. Experimental results demonstrate the method's effectiveness, which can fast synthesize novel views and 3D images with high resolution in real 3D scenes and physical simulation environments. PSNR of the virtual view is about 29.75 dB, SSIM is about 0.88, and the synthetic 8K 3D image time is about 14.41s. We believe that our fast high-resolution virtual viewpoint synthesis method can effectively improve the application of 3D light field display.

Utilizing Alike Neighbor Influenced Similarity Metric for Efficient Prediction in Collaborative Filter-Approach-Based Recommendation System

The most popular method collaborative filter approach is primarily used to handle the information overloading problem in E-Commerce. Traditionally, collaborative filtering uses ratings of similar users for predicting the target item. Similarity calculation in the sparse dataset greatly influences the predicted rating, as less count of co-rated items may degrade the performance of the collaborative filtering. However, consideration of item features to find the nearest neighbor can be a more judicious approach to increase the proportion of similar users. In this study, we offer a new paradigm for raising the rating prediction accuracy in collaborative filtering. The proposed framework uses rated items of the similar feature of the ’most’ similar individuals, instead of using the wisdom of the crowd. The reliability of the proposed framework is evaluated on the static MovieLens datasets and the experimental results corroborate our anticipations.

Advanced EEG Signal Based Min to Mean Algorithm Approach For Human Emotion Taxonomy And Mental State Analysis

With electroencephalography (EEG) brain waves alone, it is full-scale phenomena in the field of computer-brain interface DNN, CNN, and SVM have improved detection and prediction accuracy in a number of researches during the last several years. But when it comes to recognizing global reliance, both deep learning and SVM have obvious limits. Pre-processing, extraction capabilities, and network design are the most common techniques used in deep learning models today, yet they are still unable to produce reliable results in noisy and sparse datasets. Any dataset, no matter how little or large, may suffer from poor SVM performance due to overlapping target instructions and boundaries. There are many different sorts of emotions that may be classified using the particular approach employed in this research. In order to get a whole picture of a person's mental state, it is best to use a "Min of mean” proposed technique. After comparison to the referential mean, a feeling is divided into one of four emotional quadrants. The MIN Max range is used to further split the emotion into 12 subcategories based on the amount of arousal. The proposed set of rules performed better than existing methods. Research on multi-class emotion reputation has shown that, compared to more recent studies, the proposed technique may be rather strong. It is possible to analyze a person's mental health by using the emotional spectrum, which has an accuracy rate of above 90%.

Creep rupture properties of ATI 718Plus – an approach to assessing sparse data

ABSTRACT The nickel-base alloy ATI 718Plus © (UNS N07818) was introduced in 2003 with the intention of providing similar properties to Waspaloy but having the fabricability of Alloy 718. Such characteristics make it of strong interest for use in aeroengines for disk forgings, blades and bolting bar. Within power generation, the advent of higher temperatures in industrial gas turbines, and advanced supercritical steam turbines, the combination of properties and processability make it an increasingly attractive proposition for a range of applications. This paper reports an exploratory creep test programme of beyond 30 kh duration leading to an Interim ECCC Datasheet on ATI 718Plus. It demonstrates how even small volumes of test data can be assessed to produce a datasheet that can be used for preliminary design. In particular, it considers how best to determine the rupture behaviour, by including the data from unfailed tests, using both polynomial and Wilshire Equation models, with coefficients estimated by conventional regression estimation, and also with maximum likelihood methods. Additionally, creep-strain time data are modelled and extrapolated, using simplified equations developed and/or applied within the ECCC in recent work. The results confirm the temperature advantage in creep of 718Plus over Alloy 718. Recommendations of how to assess sparse datasets have also resulted from this work, leading to an expectation that new materials can be developed more quickly and with higher reliability. Nevertheless, there is no substitute for long-term multi-heat testing to prove the predicted behaviour and to examine the stability of materials in the long term.

Avatars' Social Rhythms in Online Games Indicate Their Players' Depression.

Data sets on gameplay, called digital biomarkers, contain many characteristics of game players and are associated with mental health problems. In fact, an avatar's behavior during an online game is said to be related to its player's mental health. Based on this information, we estimated the depressive state of players based on their avatar's behavioral logs. There were 3,361 participants who were players of Pigg Party, a popular Japanese online game application. In April 2020, the players logged into the Pigg Party and answered a questionnaire on depression. In May 2020, of the 3,361 participants, 658 players again logged into the Pigg Party and answered the depression questionnaire. Their responses to the questionnaire and behavioral logs of Pigg Party for April and May comprised the data sets used in the study. The data set from April showed that individuals without depression were more likely to perform activities in a 24-hour cycle than individuals with depression. Furthermore, the popular model of the time series data set trained on the April data set predicted depression in the May data set with an accuracy of around 0.3 to 0.4. The results suggest that these models are difficult to predict for sparse time series data sets, such as the study's data set, and require refinement.

Sparse Data Reconstruction, Missing Value and Multiple Imputation through Matrix Factorization

Social science approaches to missing values predict avoided, unrequested, or lost information from dense data sets, typically surveys. The authors propose a matrix factorization approach to missing data imputation that (1) identifies underlying factors to model similarities across respondents and responses and (2) regularizes across factors to reduce their overinfluence for optimal data reconstruction. This approach may enable social scientists to draw new conclusions from sparse data sets with a large number of features, for example, historical or archival sources, online surveys with high attrition rates, or data sets created from Web scraping, which confound traditional imputation techniques. The authors introduce matrix factorization techniques and detail their probabilistic interpretation, and they demonstrate these techniques’ consistency with Rubin’s multiple imputation framework. The authors show via simulations using artificial data and data from real-world subsets of the General Social Survey and National Longitudinal Study of Youth cases for which matrix factorization techniques may be preferred. These findings recommend the use of matrix factorization for data reconstruction in several settings, particularly when data are Boolean and categorical and when large proportions of the data are missing.

Estimating upper percentiles of surface water monitoring data with sparse samples

AbstractThe estimation of upper percentiles of chemical concentrations in surface water systems within sites and regions may be necessary for the assessment of potential risk to ecosystems and human health. Limited sample sizes at monitoring sites often limit the use of direct methods to estimate upper percentiles. In such cases, upper percentiles within regions within a time frame may be estimated by pooling data across sites and years, and then deriving percentile estimates from the pooled dataset. The method uses the observations resulting from either a known probability‐sampling design or a sampling design treated like one because its observations come close to matching that of a probability‐sample. These observations are then weighted to ensure that estimates are representative of a target population across all the sites within the region and the range of years in the time frame. This method of estimating upper percentiles of annual site concentration profiles is demonstrated using atrazine and validated using the monitoring data from both sparsely sampled and high‐frequency water monitoring programs, where point and interval estimates of the 90th, 95th, and 99th pooled population percentiles are provided. This method shows that the pooled data from multiple sparse datasets can be used to provide estimates of near‐peak concentrations with greater certainty, which are consistent with those generated by high‐frequency sampling monitoring programs.

Performing in-situ analytics: Mining frequent patterns from big IoT data at network edge with D-HARPP

Big IoT data is inherently distributed, high-dimensional, irregular, and sparse in nature. Fog computing model in its original form is by no means the optimal solution for mining big IoT data. However, utilizing the network edge for mining tasks, such as enabling edge and IoT devices to mine locally frequent patterns can significantly improve the mining performance. Additionally, edge devices capable of performing distributed job processing could utilize the model to the fullest. But resource poorness of edge and IoT devices needs lightweight pattern mining algorithms. This paper presents Distributed HARnessing the Power of Powersets for Mining Frequent Itemsets (D-HARPP), a spark-based distributed algorithm to mine frequent co-occurring itemsets in big IoT data. Unlike state-of-the-art distributed algorithms, D-HARPP makes a single pass over the data and does not create candidate itemsets; thus, achieves significantly better runtime and consumes the least memory. Moreover, performance of D-HARPP is not deteriorated at lower minimum support thresholds. These distinguishing characteristics make D-HARPP an optimal choice for Spark-enabled edge and IoT devices. D-HARPP has outperformed Spark-Apriori, another distributed algorithm by significant margins, both in terms of runtime and memory consumption, particularly on sparse datasets.

Review of experimental studies on Transverse Axis Crossflow Turbines

Transverse Axis Crossflow Turbines (TACTs) are a niche subset of tidal turbines. TACTs are not as well understood as the more traditional horizontal axis turbine and associated flow theory which leans heavily on advances in wind energy and marine propulsion. This paper reviews laboratory and field based experimental fluid dynamics work from the perspective of turbine performance. The available literature deviates significantly in perspective and scope since it is found that not all papers declare a full complement of parameters, thereby making it difficult to check or validate the respective results. Therefore, identifying trends amongst the variable and sparse datasets is difficult. None of the papers reviewed cite adherence to the recommended tank testing guidelines. The work reviewed analyses aspects such as mounting supports, solidity, blockage and blade support locations in isolation, but the cumulative impact of these variables is unknown. Arising from the analyses carried out as part of this review, blade loading, solidity and blockage were identified as key parameters and are the subject of planned research. Trends between solidity and blockage with tip speed ratio were identified. This paper contributes to the understanding of TACT performance through the tidal turbine performance curve and highlights the need for comprehensive physical testing and model validation.