Search

E-commerce platforms heavily rely on automatic personalized recommender systems, e.g., collaborative filtering models, to improve customer experience. Some hybrid models have been proposed recently to address the deficiency of existing models. However, their performances drop significantly when the dataset is sparse. Most of the recent works failed to fully address this shortcoming. At most, some of them only tried to alleviate the problem by considering either user side or item side content information. In this article, we propose a novel recommender model called Hybrid Variational Autoencoder (HVAE) to improve the performance on sparse datasets. Different from the existing approaches, we encode both user and item information into a latent space for semantic relevance measurement. In parallel, we utilize collaborative filtering to find the implicit factors of users and items, and combine their outputs to deliver a hybrid solution. In addition, we compare the performance of Gaussian distribution and multinomial distribution in learning the representations of the textual data. Our experiment results show that HVAE is able to significantly outperform state-of-the-art models with robust performance.

Big data has been transformed into knowledge by information systems to add value in businesses. Enterprises relying on it benefit from risk management to a certain extent. The value, however, depends on the quality of data. The quality needs to be verified before any use of the data. Specifically, measuring the quality by simulating the real life situation and even forecast it accurately turns into a hot topic. In recent years, there have been numerous researches on the measurement and assessment of data quality. These are yet to utilize a scientific computational method for the measurement and prediction. Current methods either fail to make an accurate prediction or do not consider the correlation and time sequence factors of the data. To address this, we design a model to extend machine learning technique to business applications predicting this. Firstly, we implement the model to detect data noises from a risk dataset according to an international data quality standard from banking industry and then estimate their impacts with Gaussian and Bayesian methods. Secondly, we direct sequential learning in multiple deep neural networks for the prediction with an attention mechanism. The model is experimented with various network methodologies to show the predictive power of machine learning technique and is evaluated by validation data to confirm the model effectiveness. The model is scalable to apply to any industries utilizing big data other than the banking industry.

Extracting shape-related features from a given query subsequence is a crucial preprocessing step for chart pattern matching in rule-based, template-based and hybrid pattern classification methods. The extracted features can significantly influence the accuracy of pattern recognition tasks during the data mining process. Although shape-related features are widely used for chart pattern matching in financial time series, the intrinsic properties of these features and their relationships to the patterns are rarely investigated in research community. This paper aims to formally identify shape-related features used in chart patterns and investigates their impact on chart pattern classifications in financial time series. In this paper, we describe a comprehensive analysis of 14 shape-related features which can be used to classify 41 known chart patterns in technical analysis domain. In order to evaluate their effectiveness, shape-related features are then translated into rules for chart pattern classification. We perform extensive experiments on real datasets containing historical price data of 24 stocks/indices to analyze the effectiveness of the rules. Experimental results reveal that the features put forward in this paper can be effectively used for recognizing chart patterns in financial time series. Our analysis also reveals that high-level features can be hierarchically composed from low-level features. Hierarchical composition allows construction of complex chart patterns from features identified in this paper. We hope that the features identified in this paper can be used as a reference model for the future research in chart pattern analysis.

Force-directed (FD) algorithms can be used to explore relationships in social networks, visualize money markets, and analyze transaction networks. However, FD algorithms are mainly designed for visualizing static graphs in which the topology of the networks remains constant throughout the calculation. In contrast to static graphs, nodes and edges in dynamic graphs can be added or removed as time progresses. In these situations, existing FD algorithms do not scale well, since any changes in the topology will trigger these algorithms to completely restart the entire computation. To alleviate this problem, we propose a design and implementation of five online FD algorithms to visualize dynamic graphs while maintaining their native force models. The online FD algorithms developed in this paper are able to reuse the force models of existing FD algorithms without significant modifications. To evaluate the effectiveness of the proposed approach, online FD algorithms are compared against static FD algorithms for visualizing dynamic graphs. Experimental results show that among the five algorithms evaluated, the online FD algorithm achieves the best number of edge crossings and the standard deviation of edge lengths for visualizing dynamic graphs.

Cost overruns and schedule delays are pervasive in complex projects despite the use of systems engineering and traditional project management models and tools. These disciplines can often work in isolation leading to inconsistencies in product information, tracking of design changes, and challenges in decision-making. While the literature proposes theoretical approaches to integrating these disciplines, there does not appear to be a practical approach offered. In response, this article presents a management flight simulator that represents a digital twin of set-based design, design change, knowledge, and agile project management practices. It integrates discipline-specific submodels through key linkages that are derived from the intrinsic properties of a system case study and intangible assets such as knowledge, communication, culture, and process maturity. It captures the techno-socio-economic and cultural factors involved in design change decisions and project management. The simulator provides immediate feedback on whether a change is going to help or disrupt design integrity through the monitoring of system attribute trends and cues. It also provides the impact on lifecycle management curves using a system dynamics submodel. From this feedback, several system, policy, and process levers are available within the simulator for what-if scenarios with the goal to improve product, organizational, and project performance.

In this paper, we propose a novel indoor localization system in a multi-indoor environment using cloud computing. Prior studies show that there are always concerns about how to avoid signal occlusion and interference in the single indoor environment. However, we find some general rules to support our system being immune to interference generated by occlusion in the multi-indoor environment. A convenient way is measured to deploy Bluetooth low energy devices, which mainly collect large information to assist localization. A neural network-based classification is proposed to improve localization accuracy, compared with several algorithms and their performance comparison is discussed. We also design a distributed data storage structure and establish a platform considering the storage load with Redis. Our real experimental validation shows that our system will meet the four aspects of performance requirements, which are higher accuracy, less power consumption, and increased levels of system magnitude and deployment efficiency.

Process mining is becoming an indispensable method in workflow model reconstructions, offering insights into mission critical systems. The efficacy of process mining depends on whether the underlying data mining algorithms can accurately classify or predict future events from process logs. However, exceptional events are scarce in most operational processes. Hence, the process logs generated from these processes are highly imbalanced. It is quite often that any model learned from imbalanced data tends to be overly generalized toward the normal classes but under-trained to recognize the rare classes. In this paper, we propose 3 methods to rectify this class imbalance problem. They are founded upon a meta-heuristic–swarm intelligence algorithm. The first method, and also the base of the remaining 2 methods, is Dynamic Multi-objective Rebalancing Algorithm, which considers both high accuracy and high confidence level of classification in its objective function, and it is draw upon the particle swarm optimization algorithm. The other two algorithms are hybrid methods by combining the first base method with over-sampling and under-sampling techniques. Experiments are conducted using the three above-mentioned methods to process rebalanced dataset, as well as using other classic resampling methods for comparison. According to the results, our proposed methods show satisfactory performance over other comparison methods, and we extracted meaningful decision rules from a rebalanced dataset in process mining.

© 2020, Springer Nature Switzerland AG. Sparsity and noisy labels occur inherently in real-world data. Previously, strong assumptions were made by domain experts to use their experience and expertise to select parameters for their models. Similar approach has been adopted in machine learning for hyper-parameter setting. However, these assumptions are often subjective and are not necessarily the optimal choice. To address this problem, we propose a data-driven approach to automate model parameter learning via a Bayesian nonparametric formulation. We propose hierarchical Dirichlet process mixture model (HDPMM) as a multi-task learning framework. It is used to learn the common parameters across different datasets in the same industry. In our experiments, we verified the capability of HDPMM for multi-task learning in infrastructure failure predictions. It was done by combining HDPMM with hierarchical beta process, which is our failure prediction model. In particular, multi-task learning was used to gain additional knowledge from failure records of water supply networks managed by other utility companies to improve prediction accuracy of our model. Notably, we have achieved superior accuracy for sparse predictions than previous state-of-the-art models. Moreover, we have demonstrated the capability of our proposed model in supporting preventive maintenance of critical infrastructure.

As the core of deep learning methodologies, convolutional neural network (CNN) has received wide attention in the area of image recognition. In particular, it requires very precise, accurate and fine recognition power for medical imaging processing. Numerous promising prospects of CNN applications with medical prognosis and diagnosis have been reported in the related works, and the common goal among the literature is mainly to analyze the insights from the finest details of medical images and build a more suitable model with maximum accuracy and minimum error. Thus, a novel CNN model is proposed with the characteristics of multi-view feature preprocessing and swarm-based parameter optimization. Additional information of extra features from multi-view is discovered potentially for training, and simultaneously, the most optimal set of CNN parameters are provided by our proposed leader and long-tail-based particle swarm optimization. The purpose of such a hybrid method is to achieve the highest possibility of target recognition in medical images. Preliminary experiments over cardiovascular and mammogram datasets related to heart disease prediction and breast cancer classification, respectively, are designed and conducted, and the results indicate encouraging performance compared to other existing CNN model optimization methods.

The discrete Gaussian sampler is one of the fundamental tools in implementing lattice-based cryptosystems. However, a naive discrete Gaussian sampling implementation suffers from side-channel vulnerabilities, and the existing countermeasures usually introduce significant overhead in either the running speed or the memory consumption. In this paper, we propose a fast, compact, and constant-time implementation of the binary sampling algorithm, originally introduced in the BLISS signature scheme. Our implementation adapts the Renyi divergence and the transcendental function polynomial approximation techniques. The efficiency of our scheme is independent of the standard deviation, and we show evidence that our implementations are either faster or more compact than several existing constant-time samplers. In addition, we show the performance of our implementation techniques applied to and integrated with two existing signature schemes: qTesla and Falcon. On the other hand, the convolution theorems are typically adapted to sample from larger standard deviations, by combining samples with much smaller standard deviations. As an additional contribution, we show better parameters for the convolution theorems.

Although matrix factorization and its variants have proved their effectiveness in learning user preferences, they violate the triangle inequality and may fail to capture the inner grained preference information. Thus, metric learning based models have attracted increasing interests in recommender systems. However, most of these models only measure the distance in Euclidean space, though it is known to limit the prediction performance. To address this problem, this paper proposes curvilinear collaborative metric learning (CCML) as a solution. Rather than learning from Euclidean distance, CCML measures the “path” (i.e. curvilinear distance) between the user and items. Therefore, it is able to better infer information (in geometrical representation) and user-item relations behind the observed data points. Under this setting, the relations between the user and items are not linearly reflected by the relation vector. In order to effectively learn the representation of items, we propose to fuse macro features of categories and micro representations of items to form their latent vectors. Our model can be less prone to be affected by the noise commonly presented in datasets by separately learning macro and micro features. Experimental results on eight public recommendation datasets demonstrate that CCML produces a outstanding performance compared with five competitive baselines. Finally, quantitative analysis is included to verify the efficiency and effectiveness of CCML and the macro-micro module.

Many recent relation extraction methods are proposed based on distantly supervised learning to address the issue of expensive human labelling. They automatically align relation instances from a knowledge base with unstructured text. Although these methods solve the problem of insufficient labelled data and are scalable, due to automatic labelling, they suffer from a large amount of noisy data. In this paper, we propose a relation extraction method based on a BiLSTM approach to alleviate the noise by automatic labelling. It firstly performs unsupervised pretraining for word embeddings, dependency path tokens and entity embeddings. The attention mechanism we employed in the pretraining module helps select relevant information. As pretraining denoise and encode the semantic information, the relation types can be more efficiently classified by supervised learning using less labelled training data. Experiments show that our approach significantly outperforms previous state-of-the-art baselines when labelled training data is insufficient.

Relation extraction is an important task for information extraction aiming to detect and extract semantic relationships between entity pairs in sentences. A lot of recently proposed methods for relation extraction have gained remarkable results using deep neural networks. However, a deep network usually requires expensive computational power and is time-consuming when training. To address this problem, we introduce a novel neural relation extraction model that jointly trains a seq2seq regulariser with a classification module. As the regulariser and the classification module share the encoder, the whole model is less complicated compared to novel approaches like BERT. We show that our model achieves the same or better performance than the pre-trained BERT model using sufficient labelled data in much less training time.

Locally linear embedding (LLE) is a nonlinear dimension reduction technique that only relies on the assumption of local linearity. While it is known to produce good results and is computationally efficient, it does not perform well when the observations are distorted by noises, as the fundamental assumption of local linearity becomes violated. In this work, we present a modification of locally linear embedding which is designed to handle such situations. This new modification is termed LLEAN, short for locally linear embedding with additive noise, which has been seen to perform better in the presence of noise distortion. In LLEAN, we seek to recover the noiseless data from the noisy data by exploiting the relationship between local linearity and reconstruction potential, and we then use the recovered noiseless data while performing the dimension reduction. The LLEAN algorithm includes a tuning parameter, and our work includes an automatic selection method for the tuning parameter to remove the burden from the user.

Advances in information technologies enable data to be ubiquitously generated from sensors, especially in the industrial healthcare research and application fields. The aim is to develop an adaptiv...

In future long-duration exploration missions (LDEMs), communication with mission control will be significantly delayed. Crews will often need to react to time-sensitive issues or hazards without relying on mission control for assistance. The need for a higher degree of crew autonomy without direct oversight from mission control introduces the implementation of virtual assistants (VAs) to aid the crew; however, we must first create standards and guidelines for VAs in this context. For this purpose, we have developed a VA called Daphne-AT (Anomaly Treatment) to investigate the interaction between astronauts and virtual assistants in the context of anomaly resolution related to the Environmental Control and Life Support System (ECLSS). A series of experiments will be conducted in a laboratory environment and at the Human Exploration Research Analog (HERA) at NASA Johnson Space Center (JSC) to study user's performance, situational awareness, cognitive workload, and trust in Daphne-AT. Subjects in a simulated LDEM will be given several ECLSS anomalies to identify and solve with and without the support of Daphne-AT. We based the scenarios we developed on existing ECLSS hardware and pre-existing anomaly resolution procedures from the HERA environment. Solving an anomaly requires the subject to complete the correct anomaly resolution procedure(s) for the appropriate anomaly, which could include various tasks such as swapping out component parts and activating or deactivating ECLSS systems. In experiment sessions without Daphne-AT, subjects will rely on telemetry feeds as well as background knowledge and training to solve anomalies. However, during experiment sessions with Daphne-AT, the VA will also assist subjects in detecting and diagnosing these anomalies. We conducted an initial set of experiments at Texas A&M University (TAMU) prior to those at HERA. However, the subject must complete anomaly resolution procedures during an experiment that require access to hardware and components not available at the TAMU location. To emulate the HERA ECLSS hardware elements in our laboratory at TAMU, we recreated a virtual 3D representation of HERA and its ECLSS systems using the game engine Unreal Engine 4 (UE4). This virtual model will act as an analog for subjects who are tested on TAMU's campus to complete relevant ECLSS procedures. All subsystems featured in the UE4 model are interactive and allow the user to perform steps in an anomaly resolution procedure similar to how the user would do it in HERA. The UE4 model allows the user to complete more than 20 anomaly resolution procedures, thus increasing the TAMU experiment's fidelity compared to those conducted at HERA. This paper describes the development of the UE4 model and how we used it to validate the experimental protocol implemented at HERA. The results of these experiments will inform future guidelines for VAs deployed on LDEMs.

The Global Exploration Roadmap provides a pathway for humans to leave lower Earth orbit and develop a sustained presence in cislunar space as well as the lunar surface before embarking on a long journey to Mars. During these crewed Long Duration Exploration Missions (LDEM), communication delays will limit mission control availability to support the crew during critical in-flight anomalies. In order to address this challenge, future human spaceflight operations will benefit from intelligent Virtual Assistants (VAs) capable of assisting the crew in detecting, diagnosing, and resolving emergency anomalies. This intelligent technology could increase the probability of mission success and crew safety in emergency scenarios and could improve overall crew performance while operating in the hostile environment of deep space. Daphne-AT is a VA to support crewed LDEM in the context of diagnosing and resolving in-flight anomalies related to the Environmental Control and Life Support System (ECLSS) of a spacecraft. The overall research objective is to investigate the effect of Daphne-AT on human performance, cognitive workload, situational awareness, and trust in autonomous systems. Daphne-AT will be validated using controlled laboratory experiments, and further evaluation will occur during the Human Exploration Research Analog (HERA) C6 research campaign, consisting of four 45-day missions in 2020–21. In this paper, we provide an overview of our first version of Daphne-AT (i.e., baseline version), and discuss the design of anomaly scenarios using NASA's ECLSS simulator and the experimental set up for laboratory experiments. A preliminary within-subjects (i.e., with and without Daphne-AT) experiment was conducted with five human subjects, and preliminary results indicate that the use of Daphne-AT increases performance and decreases cognitive workload in the context of ECLSS anomaly scenarios. The data also show that subjects exhibited trust in using Daphne-AT, while it improved certain aspects of their situational awareness. Our work is part of the Human Capabilities Assessments for Autonomous Missions (HCAAM) Virtual NASA Science Center of Research (VNSCOR) within the NASA Human Research Program, and the ultimate goal is to generate standards and guidelines about how to design such VAs for future LDEMs.

Decision trees are popular machine-learning classification models due to their simplicity and effectiveness. Tai et al. (ESORICS ’17) propose a privacy-preserving decision-tree evaluation protocol purely based on additive homomorphic encryption, without introducing dummy nodes for hiding the tree structure, but it runs a secure comparison for each decision node, resulting in linear complexity. Later protocols (DBSEC ’18, PETS ’19) achieve sublinear (client-side) complexity, yet the server-side path evaluation requires oblivious transfer among 2d real and dummy nodes even for a sparse tree of depth d to hide the tree structure. This paper aims for the best of both worlds and hence the most lightweight protocol to date. Our complete-tree protocol can be easily extended to the sparse-tree setting and the reusable outsourcing setting: a model owner (resp. client) can outsource the decision tree (resp. attributes) to two non-colluding servers for classifications. The outsourced extension supports multi-client joint evaluation, which is the first of its kind without using multikey fully-homomorphic encryption (TDSC ’19). We also extend our protocol for achieving privacy against malicious adversaries. Our experiments compare in various network settings our offline and online communication costs and the online computation time with the prior sublinear protocol of Tueno et al. (PETS ’19) and O(1)-round linear protocols of Kiss et al. (PETS ’19), which can be seen as garbled circuit variants of Tai et al.’s. Our protocols are shown to be desirable for IoT-like scenarios with weak clients and big-data scenarios with high-dimensional feature vectors. Info-communications Media Development Authority (IMDA) National Research Foundation (NRF) Published version

There is a growing interest in explainable machine learning methods. In our investigation, we have collected heterogeneous features from two series of YouTube video ads and seven series of Instagram picture ads to form our datasets. There are two main challenges that we found in analysing such data: i) multicollinearity and ii) infrequent common features. Due to these issues, standard estimation methods, such as OLS, Lasso, and Elastic-net, are only able to find a small number of significant features. This paper proposes a method called Significant Feature Lasso (SF-Lasso) to maximize model explainability by identifying most of the significant features that affect a target outcome (such as online video and picture ad popularity). Experiments show that SF-Lasso is able to identify much more significant features while maintaining similar prediction accuracy as what Lasso and Elastic-net can obtain. The human evaluation shows that SF-Lasso is better at identifying true features that appeal to ad viewers. We also find that the number of significant features is mainly affected by the model size (i.e., the number of active variables) and the correlations among explanatory variables.

Latent variable models have been widely adopted by recommender systems due to the advancements of their learning scalability and performance. Recent research has focused on hybrid models. However, due to the sparsity of user and/or item data, most of these proposals have convoluted model architectures and objective functions. In particular, the latter are mostly tailored for sparse data from either user or item spaces. Although it is possible to derive an analogous model for both spaces, this makes a system overly complicated. To address this problem, we propose a deep learning based latent model called Distilled Hybrid Network (DHN) with a teacher-student learning architecture. Unlike other related work that tried to better incorporate content components to improve accuracy, we instead focus on model learning optimization. To the best of our knowledge, we are the first to employ teacher-student learning architecture for recommender systems. Experiment results show that our proposed model notably outperforms state-of-the-art approaches. We also show that our proposed architecture can be applied to existing recommender models to improve their accuracies.

Big data has been transformed into knowledge by information systems to add value in businesses. Enterprises relying on it benefit from risk management to a certain extent. The value, however, depends on the quality of data. The quality needs to be verified before any use of the data. Specifically, measuring the quality by simulating the real life situation and even forecast it accurately turns into a hot topic. In recent years, there have been numerous researches on the measurement and assessment of data quality. These are yet to utilize a scientific computational method for the measurement and prediction. Current methods either fail to make an accurate prediction or do not consider the correlation and time sequence factors of the data. To address this, we design a model to extend machine learning technique to business applications predicting this. Firstly, we implement the model to detect data noises from a risk dataset according to an international data quality standard from banking industry and then estimate their impacts with Gaussian and Bayesian methods. Secondly, we direct sequential learning in multiple deep neural networks for the prediction with an attention mechanism. The model is experimented with various network methodologies to show the predictive power of machine learning technique and is evaluated by validation data to confirm the model effectiveness. The model is scalable to apply to any industries utilizing big data other than the banking industry.

Statistical divergence is widely applied in multimedia processing, basically due to regularity and interpretable features displayed in data. However, in a broader range of data realm, these advantages may no longer be feasible, and therefore a more general approach is required. In data detection, statistical divergence can be used as a similarity measurement based on collective features. In this paper, we present a collective detection technique based on statistical divergence. The technique extracts distribution similarities among data collections, and then uses the statistical divergence to detect collective anomalies. Evaluation shows that it is applicable in the real world., Comment: 6 pages, 6 figures and tables, submitted to ICME 2020

Many recently proposed relation extraction methods are based on distantly supervised learning. They use data from existing knowledge bases as training data. Although the methods solve the problem of insufficient labelled data and are highly scalable, they suffer from a large amount of incorrectly labelled data. Instead of using these distantly supervised approaches, this paper proposes an alternative relation extraction method. It firstly performs unsupervised learning to train relation embeddings by a neural network. As the relation embeddings encode the semantic information of the original sentences and their entity pairs, these embeddings can be efficiently classified by supervised learning. Since the relation embedding phase is based on unsupervised learning, labelled data is only required in the classification phase. Experiments show that our proposed approach significantly outperforms the state-of-the-art baselines when labelled training data is insufficient.

New forms of imaging modalities such as selective imaging and augmented reality eyewear, place special requirements on the optics used in these systems. This presentation reviews the properties of volume holographic materials for these applications.

The arbitrary-centered discrete Gaussian sampler is a fundamental subroutine in implementing lattice trapdoor sampling algorithms. However, existing approaches typically rely on either a fast implementation of another discrete Gaussian sampler or pre-computations with regards to some specific discrete Gaussian distributions with fixed centers and standard deviations. These approaches may only support sampling from standard deviations within a limited range, or cannot efficiently sample from arbitrary standard deviations determined on-the-fly at run-time.

Summary Genome-wide association study (GWAS) analyses, at sufficient sample sizes and power, have successfully revealed biological insights for several complex traits. RICOPILI, an open-sourced Perl-based pipeline was developed to address the challenges of rapidly processing large-scale multi-cohort GWAS studies including quality control (QC), imputation and downstream analyses. The pipeline is computationally efficient with portability to a wide range of high-performance computing environments. RICOPILI was created as the Psychiatric Genomics Consortium pipeline for GWAS and adopted by other users. The pipeline features (i) technical and genomic QC in case-control and trio cohorts, (ii) genome-wide phasing and imputation, (iv) association analysis, (v) meta-analysis, (vi) polygenic risk scoring and (vii) replication analysis. Notably, a major differentiator from other GWAS pipelines, RICOPILI leverages on automated parallelization and cluster job management approaches for rapid production of imputed genome-wide data. A comprehensive meta-analysis of simulated GWAS data has been incorporated demonstrating each step of the pipeline. This includes all the associated visualization plots, to allow ease of data interpretation and manuscript preparation. Simulated GWAS datasets are also packaged with the pipeline for user training tutorials and developer work. Availability and implementation RICOPILI has a flexible architecture to allow for ongoing development and incorporation of newer available algorithms and is adaptable to various HPC environments (QSUB, BSUB, SLURM and others). Specific links for genomic resources are either directly provided in this paper or via tutorials and external links. The central location hosting scripts and tutorials is found at this URL: https://sites.google.com/a/broadinstitute.org/RICOPILI/home Supplementary information Supplementary data are available at Bioinformatics online.

In this paper, we propose a novel technique termed as optimized swarm search-based feature selection (OS-FS), which is a swarm-type of searching function that selects an ideal subset of features for enhanced classification accuracy. In terms of gaining insights from unstructured medical based texts, sentiment prediction is becoming an increasingly crucial machine learning technique. In fact, due to its robustness and accuracy, it recently gained popularity in the medical industries. Medical text mining is well known as a fundamental data analytic for sentiment prediction. To form a high-dimensional sparse matrix, a popular preprocessing step in text mining is employed to transform medical text strings to word vectors. However, such a sparse matrix poses problems to the induction of accurate sentiment prediction model. The swarm search in our proposed OS-FS can be optimized by a new feature evaluation technique called clustering-by-coefficient-of-variation. In order to find a subset of features from all the original features from the sparse matrix, this type of feature selection has been a commonly utilized dimensionality reduction technique, and has the capability to improve accuracy of the prediction model. We implement this method based on a case scenario where 279 medical articles related to 'meaningful use functionalities on health care quality, safety, and efficiency' from a systematic review of previous medical IT literature. For this medical text mining, a multi-class of sentiments, positive, mixed-positive, neutral and negative is recognized from the document contents. Our experimental results demonstrate the superiority of OS-FS over traditional feature selection methods in literature.

Imbalanced datasets can be found in a number of fields; they are commonly regarded as big data because of their sheer volume and high attribute dimensions. As the name suggests, imbalanced big datasets come with an extremely imbalanced ratio between the amount of major class and minority class samples. Traditional methods: have been attempted but still cannot fully, effectively, and reliably solve the imbalanced class classification problem, especially when the distribution of the classes is exceedingly imbalanced. In this paper, we propose a collection of algorithms to solve the problem of imbalanced datasets in binary data classification. Most traditional methods: rebalance the imbalanced dataset merely by matching the data quantities of the two classes. Our proposed algorithms, which take the form of a suite of variants, focus on guaranteeing the credibility of the classification model and reaching the greatest possible accuracy by dynamically rebalancing the training dataset with multi-objective swarm intelligence optimisation. The new algorithms are extended from those we proposed earlier, which had a single objective – first find a set of solutions that satisfy the Kappa criterion, then search for the solution in the set that offers the highest accuracy. Two main modifications are made in the new algorithms. Multi-objective optimisation is aimed at finding a solution that satisfies several criteria at the same time, such as accuracy and identifying a list of credibility indicators. The other enhancement is the incremental operation of the multi-objective optimisation. Incremental optimisation is imperative for processing data feeds that may arrive in a streaming manner. Instead of waiting for the full data archive to be available before optimisation, incremental optimisation rebalances the data feed segment by segment on the fly. The experimental results from the suite of proposed algorithms show that they can effectively attain better and more stable performances from the classification model and are accompanied by much greater credibility than the other five traditional methods when imbalanced datasets are used as training datasets for inducing a classifier.

Decades-long experimental databases become accessible in global earthquake engineering community. Yet, complex interactions of a multitude of variables pose formidable challenges to data-driven res...

Data clustering is one of the most popular branches of machine learning and data analysis. Partitioning-based type of clustering algorithms, such as K-means, is prone to the problem of producing a set of clusters that is far from perfect due to its probabilistic nature. The clustering process starts with some random partitions at the beginning, and then it attempts to improve the partitions progressively. Different initial partitions can result in different final clusters. Trying through all the possible candidate clusters for the perfect result is computationally expensive. Meta-heuristic algorithm aims to search for global optimum in high-dimensional problems. Meta-heuristic algorithm has been successfully implemented on data clustering problems seeking a near optimal solution in terms of quality of the resultant clusters. In this paper, a new meta-heuristic search method named elephant search algorithm (ESA) is proposed to integrate into K-means, forming a new data clustering algorithm, namely C-ESA. The advantage of C-ESA is its dual features of (i) evolutionary operations and (ii) balance of local intensification and global exploration. The results by C-ESA are compared with classical clustering algorithms including K-means, DBSCAN, and GMM-EM. C-ESA is shown to outperform the other algorithms in terms of clustering accuracy via a computer simulation. C-ESA is also implemented on time series clustering compared with classical algorithms K-means, Fuzzy C-means and classical meta-heuristic algorithm PSO. C-ESA outperforms the other algorithms in term of clustering accuracy. C-ESA is still comparable compared with state of art time series clustering algorithm K-shape.

In this paper, a novel meta-heuristic search algorithm inspired by rhinoceros’ natural behaviour is proposed, namely rhinoceros search algorithm (RSA). Similar to our earlier version called elephant search algorithm, RSA simplifies certain habitual characteristics of rhinoceros and stream-lines the search operations, thereby reducing the number of operational parameters required to configure the model. Via computer simulation, it is shown that RSA is able to outperform certain classical meta-heuristic algorithms. Different dimensions of optimization problems are tested, and good results are observed by RSA. The RSA is also implemented on permutation flow-shop scheduling problem (PFSP) with some representation method. Four different problem scales are used. Compared with partible swarm optimization (PSO) on PFSP, the RSA outperforms PSO on different problem scales with a 3% improvement.

With the popularity and affordability of ZigBee wireless sensor technology, IoT-based smart controlling system for home appliances becomes prevalent for smart home applications. From the data analytics point of view, one important objective from analyzing such IoT data is to gain insights from the energy consumption patterns, thereby trying to fine-tune the energy efficiency of the appliance usage. The data analytics usually functions at the back-end crunching over a large archive of big data accumulated over time for learning the overall pattern from the sensor data feeds. The other objective of the analytics, which may often be more crucial, is to predict and identify whether an abnormal consumption event is about to happen. For example, a sudden draw of energy that leads to hot spot in the power grid in a city, or black-out at home. This dynamic prediction is usually done at the operational level, with moving data stream, by data stream mining methods . In this paper, an improved version of very fast decision tree (VFDT) is proposed, which learns from misclassified results for the sake of filtering the noisy data from learning and maintaining sharp classification accuracy of the induced prediction model. Specifically, a new technique called misclassified recall (MR), which is a pre-processing step for self-rectifying misclassified instances, is formulated. In energy data prediction, most misclassified instances are due to data transmission errors or faulty devices. The former case happens intermittently, and the errors from the latter cause may persist for a long time. By caching up the data at the MR pre-processor, the one-pass online model learning can be effectively shielded in case of intermitting problems at the wireless sensor network; likewise the stored data could be investigated afterwards should the problem persist for long. Simulation experiments over a dataset about predicting exceptional appliances energy use in a low energy building are conducted. The reported results validate the efficacy of the new methodology VFDT + MR, in comparison to a collection of popular data stream mining algorithms from the literature.

Dynamic group optimization has recently appeared as a novel algorithm developed to mimic animal and human socialising behaviours. Although the algorithm strongly lends itself to exploration and exploitation, it has two main drawbacks. The first is that the greedy strategy, used in the dynamic group optimization algorithm, guarantees to evolve a generation of solutions without deteriorating than the previous generation but decreases population diversity and limit searching ability. The second is that most information for updating populations is obtained from companions within each group, which leads to premature convergence and deteriorated mutation operators. The dynamic group optimization with a mean–variance search framework is proposed to overcome these two drawbacks, an improved algorithm with a proportioned mean solution generator and a mean–variance Gaussian mutation. The new proportioned mean solution generator solutions do not only consider their group but also are affected by the current solution and global situation. The mean–variance Gaussian mutation takes advantage of information from all group heads, not solely concentrating on information from the best solution or one group. The experimental results on public benchmark test suites show that the proposed algorithm is effective and efficient. In addition, comparative results of engineering problems in welded beam design show the promise of our algorithms for real-world applications.

A new paradigm of IoT monitoring using sonar sensors and microphones is studied as a contactless alternative to the traditional devices for sleep disorders in this paper. A pair of sonar sensors are used to measure the distance from the bottom and side of the body to the bedside respectively, and the basic posture of the human body during sleep can be inferred by using machine learning. When a person sleeps still, two streams of sonar signals from two adjacent sides remain unchanged. Any movement would disrupt the stationary sonar streams. Hence it could be detected as change of posture from lying still. Different sonar patterns could be recognized as specific sleeping postures by some non-linear machine learning model. This simple and novel solution could potentially be used as an alternative or supplementary to video analytics which can feedback to the user about their sleeping pattern. A new data transformation method namely Kennard-stone Balance (KSB) algorithm is also proposed for simplifying the data streams and enhancing the accuracy of the machine learning model. Simulation results show the feasibility of this sonar method, and KSB is able to improve the pattern recognition performance.

Service recommendation systems have been trying to utilize context-aware information to recommend services that better meet the needs of the service consumers. However, current context-aware service recommendation techniques are mainly based on individual intelligence or the local knowledge of users, and do not take into consideration the common knowledge among different users. To address this, recent research has attempted to use role-based approaches to recommend services to other members within the same context group. However, these proposed algorithms are inefficient and may not scale to cope with the large amount of mobile traffic in the real-world. This paper proposes novel algorithms with better runtime complexity, and further extends them to a MapReduce style to take advantage of popular distributed computing platforms. Experiments running on a medium-sized high performance computing cluster demonstrate that our proposed algorithms outperform previous work in runtime complexity and scalability.

The number of sensors deployed around the world is growing at a rapid pace when we are moving towards the Internet of Things (IoT). The widespread deployment of these sensors represents significant financial investment and technical achievement. These sensors continuously generate enormous amounts of data which is capable of supporting an almost unlimited set of high value proposition applications for users. Given that, effectively and efficiently searching and selecting the most related sensors of a user’s interest has recently become a crucial challenge. In this paper, inspired by ant clustering algorithm, we propose an effective context-aware method to cluster sensors in the form of Sensor Semantic Overlay Networks (SSONs) in which sensors with similar context information are gathered into one cluster. Firstly, sensors are grouped based on their types to create SSONs. Then, our meta-heuristic algorithm called AntClust has been performed to cluster sensors using their context information. Furthermore, useful adjustments have been applied to reduce the cost of sensor search process and an adaptive strategy is proposed to maintain the performance against dynamicity in the IoT environment. Experiments show the scalability and adaptability of AntClust in clustering sensors. It is significantly faster on sensor search when compared with other approaches.

Most state-of-the-art service selection approaches assume user preferences can be provided by the target user with sufficient precision and ignore historical service usage data for all users. It is desirable for ordinary users to possess a new service selection approach that can recommend satisfactory services to them even when their service selection preferences are specified imprecisely in terms of vagueness, inaccuracy, and incompleteness. This paper proposes a novel service selection approach that resolves the imprecise characteristics of user preferences and can recommend satisfactory services for users with varying cognitive levels in terms of service experience. The proposed service selection approach is comprised of four major tasks: 1) employ user-friendly linguistic variables to collect apparent user preferences (AUP) and convert the linguistic variables to standardized fuzzy weights as AUP weights; 2) evaluate all users’ respective cognitive levels for the target service type and obtain the cognitive level threshold for that type of services; 3) adjust the AUP weights based on the calculated cognitive levels and the threshold, and supplement the potential user preferences weights; and 4) prioritize candidate services per a user satisfaction maximization objective. In-depth comparative experimental evaluations were performed using two real-world datasets. The results show that our service selection model outperforms three other representative ones and could provide a stable and reliable selection of services for the users with low service cognitive levels.

Devices embedded with position tracking facilities are now widely available, such as smartphones, smartwatches, vehicle location trackers, etc. However, data mining and advanced analytics are rarely bundled with these devices that limits their utility. In this paper, we present the design of a generic, programmable position tracking platform, namely CQtracker. In particular, this platform is incorporated with a cloud-based engine of advanced analytics. CQtracker is constructed based on a concept of system-of-systems service architecture to deliver data-system-as-a-service. It is designed for the consumption by a variety of spatio-temporal applications. Spatio-temporal data exhibit strong heterogeneous patterns, data sparseness and distribution skewness. Hence, they are difficult to analyze. CQtracker reveals relationships and structures from these data by self-regularized time-varying dynamic Bayesian networks. In addition, a Bayesian parameter estimation approach is applied to an epidemic model for outbreak predictions. Sample applications are presented in this paper, in which CQtracker successfully reveals the evolution of time-varying structures from traffic trajectories.

Smart city is an emerging research field nowadays, with emphasis of using big data to enhance citizens' quality of life. One of the prevalent smart city projects is to use big traffic data collected from road users over time, for road planning, traffic light scheduling, traffic jam relief, and public security. In particular, being able to know a road user's current location and predict his/her next move is important in today's intelligent transportation systems. Trajectory prediction has become a prudential research study direction, by which many algorithms have been published before. In this paper, we present a simple probabilistic model which predicts road users' next locations based on the "concept of segments" abstracted from historical trails which the users have taken and accumulated over time in some data archive. Given a trajectory and a current location, the road user's next move in terms of road direction can be predicted at the junction. It is found that each road user would have his/her unique travel pattern hidden in the aggregate big traffic data. These patterns could be modeled from connected segments for simplicity. With the longer the trail and more frequently this trail was travelled, the more accurate that the next turn can be predicted. Simulation experiment was conducted based on summing up the segments from empirical trajectory data that was used in trajectory data mining by Microsoft. The results of our alternative model in contrast to the state of the arts demonstrated good efficacy.

Record Matching refers to identifying pairs of records that relate to the same entities across different data sources. In many applications of data mining, record matching is usually associated to quadratic complexity. In practice, the number of non-matching record pairs always far exceeds the number of matching pairs, and this is called imbalance problem. Blocking is a technique of data reduction, which can filter unlikely matching pairs before record matching. However, for big data there is no fast and effective blocking algorithm yet. In this paper, we report on big data infrastructure to improve efficiency of blocking. Our approach runs blocking process independently and distributedly on the partitions of whole data. To improve efficiency, we adopt a probabilistic technique to balance the speed and the effect of the algorithm that we proposed for distributed blocking. Our experimental analysis endorses the superiority of our technique and shows its novel scalability.

Location-based services on GPS-enabled smartphones are undergoing strong growth. Capitalizing on the popularity of this geo-location social media, a mobile app called Foursquare is developed to recommend its users places where they may be interested in, to travel from their current proximities. Such location data, in the form of check-ins by Foursquare, have huge business potentials including marketing, advertising and consumers' behaviors analysis. Many researchers from both academia and industries are seriously looking into this location-based big data which comes in high velocity (with millions of users and frequent geo-tagging), and wide variety (with potentially many meta-data and associations), accumulating into a huge volume. One of the fundamental analytics in such big data is to guess which check-in locations a user would move to, as a prerequisite for sequential mining and other lifestyle pattern analysis. This paper reports a novel, but simple big data analytic by sampling a portion of location data for predicting the next check-in locations. This proposed analytic does not need every individual user's history path and ID to match the history path of the current user in the database in order to infer a prediction. We show by a simulation experiment based on a Foursquare dataset that a minimum of two pairs of coordinates are required to provide a prediction. Several variables such as segment lengths, number of check-ins, and time factors are investigated in the experiment in relation to the prediction accuracy.

A remarkable recent discovery in machine learning has been that deep neural networks can achieve impressive performance (in terms of both lower training error and higher generalization capacity) in the regime where they are massively over-parameterized. Consequently, over the past year, the community has devoted growing interest in analyzing optimization and generalization properties of over-parameterized networks, and several breakthrough works have led to important theoretical progress. However, the majority of existing work only applies to supervised learning scenarios and hence are limited to settings such as classification and regression. In contrast, the role of over-parameterization in the unsupervised setting has gained far less attention. In this paper, we study the gradient dynamics of two-layer over-parameterized autoencoders with ReLU activation. We make very few assumptions about the given training dataset (other than mild non-degeneracy conditions). Starting from a randomly initialized autoencoder network, we rigorously prove the linear convergence of gradient descent in two learning regimes, namely: (i) the weakly-trained regime where only the encoder is trained, and (ii) the jointly-trained regime where both the encoder and the decoder are trained. Our results indicate the considerable benefits of joint training over weak training for finding global optima, achieving a dramatic decrease in the required level of over-parameterization. We also analyze the case of weight-tied autoencoders (which is a commonly used architectural choice in practical settings) and prove that in the over-parameterized setting, training such networks from randomly initialized points leads to certain unexpected degeneracies., Added Sections 3.2 and 3.4 on inductive biases. Fixed an error in deriving the neural tangent kernel in Section 3.3

One of the leading causes of disability is rheumatoid arthritis (RA), especially that of the hand, the metacarpophalangeal (MCP) joint being affected the most. Current single piece prostheses are designed to allow for motion in flexion and extension, but little effort is put towards motion in abduction and adduction. The objective of this study is to determine the effects of web (hinge) joint cross-sectional geometry on stress, strain, shear strain, strain energy, and reaction moment magnitude for an MCP joint that is subject to not only flexion and extension but also abduction and adduction. Using finger bone dimensions from the literature, geometry was produced in ANSYS finite element software. The geometry was assigned a hyperelastic material constitutive model, making the analysis nonlinear. The cross-sectional shape of the hinge was controlled by ellipse dimensions, one for thickness and one for width. Motion boundary conditions were applied to the distal portion of the model resulting in bending at the hinge. The study showed that for flexion/extension motion the von Mises (equivalent) stress, shear strain, and equivalent strain are linearly proportional to the thickness but inversely proportional to the width. The reaction moment and strain energy are linearly proportional to the thickness but exponentially proportional to the width. For motion in abduction and adduction the behavior is opposite; the width acting as the thickness does in flexion/extension motion and the thickness acting as the width does in flexion/extension motion. It was also seen that high levels of shear strain develop on the palmer side of the model, indicating that failure has the most potential to occur in that area.

We introduce MatRiCT, an efficient RingCT protocol for blockchain confidential transactions, whose security is based on "post-quantum'' (module) lattice assumptions. The proof length of the protocol is around two orders of magnitude shorter than the existing post-quantum proposal, and scales efficiently to large anonymity sets, unlike the existing proposal. Further, we provide the first full implementation of a post-quantum RingCT, demonstrating the practicality of our scheme. In particular, a typical transaction can be generated in a fraction of a second and verified in about 23 ms on a standard PC. Moreover, we show how our scheme can be extended to provide auditability, where a user can select a particular authority from a set of authorities to reveal her identity. The user also has the ability to select no auditing and all these auditing options may co-exist in the same environment. The key ingredients, introduced in this work, of MatRiCT are 1) the shortest to date scalable ring signature from standard lattice assumptions with no Gaussian sampling required, 2) a novel balance zero-knowledge proof and 3) a novel extractable commitment scheme from (module) lattices. We believe these ingredients to be of independent interest for other privacy-preserving applications such as secure e-voting. Despite allowing 64-bit precision for transaction amounts, our new balance proof, and thus our protocol, does not require a range proof on a wide range (such as 32- or 64-bit ranges), which has been a major obstacle against efficient lattice-based solutions. Further, we provide new formal definitions for RingCT-like protocols, where the real-world blockchain setting is captured more closely. The definitions are applicable in a generic setting, and thus are believed to contribute to the development of future confidential transaction protocols in general (not only in the lattice setting).

While time-series analysis is commonly used in financial forecasting, a key source of market-sentiments is often omitted. Financial news is known to be making persuasive impact on the markets. Without considering this additional source of signals, only sub-optimal predictions can be made. This paper proposes a notion of sentiment-of-topic (SoT) to address the problem. It is achieved by considering sentiment-linked topics, which are retrieved from time-series with heterogeneous dimensions (i.e., numbers and texts). Using this approach, we successfully improve the prediction accuracy of a proprietary trade recommendation platform. Different from traditional sentiment analysis and unsupervised topic modeling methods, topics associated with different sentiment levels are used to quantify market conditions. In particular, sentiment levels are learned from historical market performances and commentaries instead of using subjective interpretations of human expressions. By capturing the domain knowledge of respective industries and markets, an impressive double-digit improvement in portfolio return is obtained as shown in our experiments.