Millions Of Instances Research Articles

A distributed evolutionary based instance selection algorithm for big data using Apache Spark

Instance selection is an important preprocessing technology in data mining and machine learning. In this paper, we proposed a novel evolutionary based instance selection algorithm for big data. First, we defined a coarse granularity chromosome structure to reduce the size of search space and costs of chromosome operations(recombination and mutation, etc.). Then a stratified evolution strategy was proposed to remove the hyper parameter in classic fitness function and achieve precise control over the reduction ratio of instances. Finally, a sampling-based fitness function was proposed to reduce the time complexity. Experimental results shown that our new algorithm is efficient to complete the instance selection task on data set with millions of instances in minutes-level. The 10-fold cross-validation also proved that the selection results on many datasets have high nearest neighbor classification accuracy.

Role of Artificial Intelligence (AI) in Changing Consumer Buying Behaviour

This case study examines how positively Artificial Intelligence (AI) has affected marketers' capacity to analyse and comprehend customer behaviour. Marketers are studying how people act online in order to improve the effectiveness of their marketing plans and strategies. Given the enormous amount of data presently available and the frequency of data breaches, AI might be the solution. By studying millions of instances, an image recognition technology may be able to detect and classify objects in photographs. A chatbot can learn to have realistic conversations with people by being shown instances of text chats. To keep ahead of the competition, businesses must now leverage the greatest AI talent. It can shed light on each stage of the customer journey and help marketers better understand the driving forces behind consumer behaviour. When done correctly, it can result in positive outcomes including a higher client lifetime value and stronger client connections. Marketing teams may use AI to decipher massive amounts of data in order to utilise the information and identify the target market. They can utilise it to create user-centred sales funnels and base their marketing strategies on them. More traffic is ultimately generated, which is positive for marketing teams looking to convert visitors. Technology like artificial intelligence (AI) is changing the way we view and comprehend marketing.

Alpha-divergence minimization for deep Gaussian processes

This paper proposes the minimization of α-divergences for approximate inference in the context of deep Gaussian processes (DGPs). The proposed method can be considered as a generalization of variational inference (VI) and expectation propagation (EP), two previously used methods for approximate inference in DGPs. Both VI and EP are based on the minimization of the Kullback-Leibler divergence. The proposed method is based on a scalable version of power expectation propagation, a method that introduces an extra parameter α that specifies the targeted α-divergence to be optimized. In particular, such a method can recover the VI solution when α→0 and the EP solution when α→1. An exhaustive experimental evaluation shows that the minimization of α-divergences via the proposed method is feasible in DGPs and that choosing intermediate values of the α parameter between 0 and 1 can give better results in some problems. This means that one can improve the results of VI and EP when training DGPs. Importantly, the proposed method allows for stochastic optimization techniques, making it able to address datasets with several millions of instances.

Improved meta‐heuristic algorithm for selecting optimal features: A big data classification model

AbstractMany fields function with large databases constitute a high number of features. Feature selection strategies seek to exclude the features that are distracting, repetitive, or unnecessary, as they can degrade the classification results. Existing approaches lack the scalability needed to handle the datasets with millions of instances and they do not obtain favorable results in a timely manner. This study uses a unique feature selection approach based on an upgraded optimization model and deep machine learning‐based data classification. “(a) Feature extraction, (b) optimal feature selection, and (c) classification” are the three stages of the proposed model. Initially, the extracted big‐datasets are efficiently handled by the parallel pool map‐reduce architecture. Several features from the input big‐data are extracted using feature extraction (FE) approaches such as the suggested Tri‐Kernel principal component analysis (TK‐PCA), linear discriminant analysis, and linear square regression. Furthermore, the data obtained characteristics may contain data that is irrelevant, out‐of‐date, or noisy. The computing cost rises due to the larger feature space. As a result, the best features are selected using a new optimization technique known as Levy Adapted SLnO (LA‐SLnO), which is a superior variant of the original SLnO algorithm. This selection of appropriate features improves the classification accuracy. For classification, Convolutional Neural Network is used in this work. Finally, a comparative evaluation is undergone to validate the efficiency of the proposed model.

Parallel and accurate k‐means algorithm on CPU‐GPU architectures for spectral clustering

Summaryk‐Means is a standard algorithm for clustering data. It constitutes generally the final step in a more complex chain of high‐quality spectral clustering. However, this chain suffers from lack of scalability when addressing large datasets. This can be overcome by applying also the k‐means algorithm as a preprocessing task to reduce the input data instances. We propose parallel optimization techniques for the k‐means algorithm on CPU and GPU. Particularly we use a two‐step summation method with package processing to handle the effect of rounding errors that may occur during the phase of updating cluster centroids. Our experiments on synthetic and real‐world datasets containing millions of instances exhibit a speedup up to 7 for the k‐means iteration time on GPU versus 20/40 CPU threads using AVX units, and achieve double‐precision accuracy with single‐precision computations.

Multimodal Data Fusion in High-Dimensional Heterogeneous Datasets Via Generative Models

The commonly used latent space embedding techniques, such as Principal Component Analysis, Factor Analysis, and manifold learning techniques, are typically used for learning effective representations of homogeneous data. However, they do not readily extend to heterogeneous data that are a combination of numerical and categorical variables, e.g., arising from linked GPS and text data. In this paper, we are interested in learning probabilistic generative models from high-dimensional heterogeneous data in an unsupervised fashion. The learned generative model provides latent unified representations that capture the factors common to the multiple dimensions of the data, and thus enable fusing multimodal data for various machine learning tasks. Following a Bayesian approach, we propose a general framework that combines disparate data types through the natural parameterization of the exponential family of distributions. To scale the model inference to millions of instances with thousands of features, we use the Laplace-Bernstein approximation for posterior computations involving nonlinear link functions. The proposed algorithm is presented in detail for the commonly encountered heterogeneous datasets with real-valued (Gaussian) and categorical (multinomial) features. Experiments on two high-dimensional and heterogeneous datasets (NYC Taxi and MovieLens-10M) demonstrate the scalability and competitive performance of the proposed algorithm on different machine learning tasks such as anomaly detection, data imputation, and recommender systems.

SI(FS)[formula omitted]: Fast simultaneous instance and feature selection for datasets with many features

Data reduction is becoming increasingly relevant due to the enormous amounts of data that are constantly being produced in many fields of research. Instance selection is one of the most widely used methods for this task. At the same time, most recent pattern recognition problems involve highly complex datasets with a large number of possible explanatory variables. For many reasons, this abundance of variables significantly hinders classification and recognition tasks. There are efficiency issues, too, because the speed of many classification algorithms is greatly improved when the complexity of the data is reduced. Thus, feature selection is also a widely used method for data reduction and for gaining an understanding of feature information.Although most methods address instance and feature selection separately, the two problems are interwoven, and benefits are expected from performing these two tasks jointly. However, few algorithms have been proposed for simultaneously addressing the tasks of instance and feature selection. Furthermore, most of those methods are based on complex heuristics that are very difficult to scale up even to moderately large datasets. This paper proposes a new algorithm for dealing with many instances and many features simultaneously by performing joint instance and feature selection using a simple heuristic search and several scaling-up mechanisms that can be successfully applied to datasets with millions of features and instances.In the proposed method, a forward selection search is performed in the feature space jointly with the application of standard instance selection in a constructive subspace built stepwise. Several simplifications are adopted in the search to obtain a scalable method. An extensive comparison using 95 large datasets shows the usefulness of our method and its ability to deal with millions of instances and features simultaneously. The method is able to obtain better classification performance results than state-of-the-art approaches while achieving considerable data reduction.

Parallel Fractional Hot-Deck Imputation and Variance Estimation for Big Incomplete Data Curing

The fractional hot-deck imputation (FHDI) is a general-purpose, assumption-free imputation method for handling multivariate missing data by filling each missing item with multiple observed values without resorting to artificially created values. The corresponding R package FHDI J. Im, I. Cho, and J. K. Kim, “An R package for fractional hot deck imputation,” <i>R J.</i> , vol. 10, no. 1, pp. 140–154, 2018 holds generality and efficiency, but it is not adequate for tackling big incomplete data due to the requirement of excessive memory and long running time. As a first step to tackle big incomplete data by leveraging the FHDI, we developed a new version of a parallel fractional hot-deck imputation (named as P-FHDI) program suitable for curing large incomplete datasets. Results show a favorable speedup when the P-FHDI is applied to big datasets with up to millions of instances or 10,000 of variables. This paper explains the detailed parallel algorithms of the P-FHDI for large instances (big- <inline-formula><tex-math notation="LaTeX">$n$</tex-math></inline-formula> ) or high-dimensionality (big- <inline-formula><tex-math notation="LaTeX">$p$</tex-math></inline-formula> ) datasets and confirms the favorable scalability. The proposed program inherits all the advantages of the serial FHDI and enables a parallel variance estimation, which will benefit a broad audience in science and engineering.

Sample size determination for biomedical big data with limited labels

The era of big data has produced vast amounts of information that can be used to build machine learning models. In many cases, however, there is a point where adding more data only marginally increases model performance. This is especially important for scenarios of limited labeled data, as annotation can be expensive and time consuming. If the required sample size for accurate model performance can be determined early, then resources can be allocated appropriately to minimize time and cost. In this study, we explore sample size determination methods for four real-world biomedical datasets, spanning genomics, proteomics, electronic health records, and insurance claims data, all with millions of instances each and<2% class ratio. The methods used involve approximating a learning curve for a large amount of data using a small amount of data. We evaluate an existing method that fits an inverse power law model to a small learning curve and introduce a novel semi-supervised method that utilizes the large amount of unlabeled data for estimating a learning curve. We find that the inverse power law method is applicable to big data, while the semi-supervised method can be better at detecting convergence. To the best of our knowledge, this is the first study to apply an inverse power law curve fitting method to big data with limited labels and compare it to a semi-supervised approach.

Alpha divergence minimization in multi-class Gaussian process classification

This paper analyzes the minimization of α-divergences in the context of multi-class Gaussian process classification. For this task, several methods are explored, including memory and computationally efficient variants of the Power Expectation Propagation algorithm, which allow for efficient training using stochastic gradients and mini-batches. When these methods are used for training, very large datasets (several millions of instances) can be considered. The proposed methods are also very general as they can interpolate between other popular approaches for approximate inference based on Expectation Propagation (EP) (α → 1) and Variational Bayes (VB) (α → 0) simply by varying the α parameter. An exhaustive empirical evaluation analyzes the generalization properties of each of the proposed methods for different values of the α parameter. The results obtained show that one can do better than EP and VB by considering intermediate values of α.

Developments in Pyrotechnic-Assisted Fuses [Happenings

Every electric power engineer has some familiarity with electrical faults and technologies for fault mitigation. The basic problem presented itself to the earliest workers in electric technology, and the basic solutions have been present essentially from the beginning, and many are used with great effect to this day. In some applications where essentially the same protections are required in millions of instances per year, very cost-effective solutions are common. Mainstream automobiles (i.e., setting aside for the moment battery-electric and hybrid-electric vehicles), for example, are largely protected by simple electrical fuses, which are technically effective at very low cost. Residential electrical systems at one time used fuses, too, but these have mostly been supplanted by resettable electromechanical circuit breakers.

On building and publishing Linked Open Schema from social Web sites

Schema-level knowledge is important for different semantic applications, such as reasoning, data integration and question answering. Compared with billions of triples describing millions of instances, current Linking Open Data has only a limited number of triples representing schema-level knowledge. To facilitate multilingual schema-level knowledge mining, we propose a general approach to learn Linked Open Schema (LOS) in different languages from social Web sites, which contain rich sources (i.e. taxonomies composed of categories and folksonomies consisting of tags) for mining large-scale schema-level knowledge. The core part of the proposed approach is a semi-supervised learning method integrating rules to capture equal, subClassOf and relate relations among the collected categories and tags. We respectively apply the proposed approach to the selected English social Web sites and the Chinese ones, resulting in an English LOS and a Chinese LOS. We publish the English LOS and the Chinese one as open data on the Web with three access levels, i.e. data dump, lookup service and SPARQL endpoint. Experimental results show the high accuracy of the relations in the English LOS and the Chinese one. Compared with DBpedia, Yago, BabelNet, and Freebase, both the English LOS and the Chinese one not only have large-scale concepts, but also contain the largest number of subClassOf relations.

Isolation‐based anomaly detection using nearest‐neighbor ensembles

AbstractThe first successful isolation‐based anomaly detector, ie, iForest, uses trees as a means to perform isolation. Although it has been shown to have advantages over existing anomaly detectors, we have identified 4 weaknesses, ie, its inability to detect local anomalies, anomalies with a high percentage of irrelevant attributes, anomalies that are masked by axis‐parallel clusters, and anomalies in multimodal data sets.To overcome these weaknesses, this paper shows that an alternative isolation mechanism is required and thus presents iNNE or isolation using Nearest Neighbor Ensemble.Although relying on nearest neighbors, iNNE runs significantly faster than the existing nearest neighbor–based methods such as the local outlier factor, especially in data sets having thousands of dimensions or millions of instances. This is because the proposed method has linear time complexity and constant space complexity.

On Building and Publishing Linked Open Schema from Social Web Sites

Schema-level knowledge is important for different semantic applications, such as reasoning, data integration and question answering. Compared with billions of triples describing millions of instances, current Linking Open Data has only a limited number of triples representing schema-level knowledge. To facilitate multilingual schema-level knowledge mining, we propose a general approach to learn Linked Open Schema (LOS) in different languages from social Web sites, which contain rich sources (i.e. taxonomies composed of categories and folksonomies consisting of tags) for mining large-scale schema-level knowledge. The core part of the proposed approach is a semi-supervised learning method integrating rules to capture equal, subClassOf and relate relations among the collected categories and tags. We respectively apply the proposed approach to the selected English social Web sites and the Chinese ones, resulting in an English LOS and a Chinese LOS.We publish the English LOS and the Chinese one as open data on the Web with three access levels, i.e. data dump, lookup service and SPARQL endpoint. Experimental results show the high accuracy of the relations in the English LOS and the Chinese one. Compared with DBpedia, Yago, BabelNet, and Freebase, both the English LOS and the Chinese one not only have large-scale concepts, but also contain the largest number of subClassOf relations.

Embedded Feature Selection Method for a Network-Level Behavioural Analysis Detection Model

Feature selection in network-level behavioural analysis studies is used to represent the network datasets of a monitored space. However, recent studies have shown that current behavioural analysis methods at the network-level have several issues. The reduction of millions of instances, disregarded parameters, removed similarities of most of the traffic flows to reduce information noise, insufficient number of optimised features and ignore instances which are not an entity are amongst the other issue that have been identified as the main issues contributing to the inability to predict zero-day attacks. Therefore, this paper aims to select the optimal features that will improve the prediction and behavioural analysis. The training dataset will be trained to use the embedded feature selection method which incorporates both the filter and wrapper method. Correlation coefficient, r and weighted score, wj will be used. The accepted or selected features will be optimised uses Beta distribution functions, β, to find its maximum likelihood, Ɩmax. The final selected features will be trained by the Bayesian Network classifier and tested through several testing datasets. Finally, this method was compared to several other feature selection methods. Final results show the proposed selection method’s performance against other datasets consistently outperform other methods.

IMPACT OF ARTIFICIAL INTELLIGENCE ON HUMAN RIGHTS IN INDIA: ACRITICAL STUDY

A computer system with artificial intelligence (AI) can carry out activities that typically require human intelligence. These artificial intelligence systems are frequently powered by machine learning, occasionally by deep learning, and occasionally by really dull stuff like rules. This results from learning, which is acquiring the knowledge and guidelines for using the data. It has grown to be quite popular and necessary as a result of data-based service sectors. The ability of a computer or robot operated by a computer to carry out tasks usually performed by intelligent beings. Massive amounts of labeled training data are ingested by AI systems, which then analyze the data for correlations and patterns before using these patterns to forecast future states. By analyzing millions of instances, an image recognition tool can learn to recognize and describe items in photographs, just as a chat-bot that is fed examples of text chats can learn to make lifelike interactions with humans. AI has the potential to help societies overcome some of the biggest problems of our day. It also presents a number of threats to the same civilization, though. What are the significant concerns that AI poses to society and how do they affect human rights? This paper is about issues including the effects of artificial intelligence (AI) on the Indian legal system and its challenges.

Online estimation of discrete, continuous, and conditional joint densities using classifier chains

We address the problem of estimating discrete, continuous, and conditional joint densities online, i.e., the algorithm is only provided the current example and its current estimate for its update. The family of proposed online density estimators, estimation of densities online (EDO), uses classifier chains to model dependencies among features, where each classifier in the chain estimates the probability of one particular feature. Because a single chain may not provide a reliable estimate, we also consider ensembles of classifier chains and ensembles of weighted classifier chains. For all density estimators, we provide consistency proofs and propose algorithms to perform certain inference tasks. The empirical evaluation of the estimators is conducted in several experiments and on datasets of up to several millions of instances. In the discrete case, we compare our estimators to density estimates computed by Bayesian structure learners. In the continuous case, we compare them to a state-of-the-art online density estimator. Our experiments demonstrate that, even though designed to work online, EDO delivers estimators of competitive accuracy compared to other density estimators (batch Bayesian structure learners on discrete datasets and the state-of-the-art online density estimator on continuous datasets). Besides achieving similar performance in these cases, EDO is also able to estimate densities with mixed types of variables, i.e., discrete and continuous random variables.

SMOTE-GPU: Big Data preprocessing on commodity hardware for imbalanced classification

Nowadays, it is usual to work with large amounts of data since our capacity of collecting and storing information has increased significantly. The extraction of knowledge from these scenarios is commonly known as “Big Data,” and it is performed on large clusters with MapReduce platforms. Imbalanced classification poses a problem both in traditional and Big Data learning scenarios. Data sampling is one of the ways that allows to improve the performance on imbalanced problems. A commodity hardware-based method for Big Data problems can offload these computations from the expensive and highly demanded hardware that MapReduce platforms require. The characteristics of some sampling methods make them suitable to be adapted to commodity hardware, taking advantage of the parallel computation capabilities of graphics processing units. SMOTE is one of the most popular oversampling methods which is based on the nearest neighbor rule. The proposed SMOTE-GPU efficiently handles large datasets (several millions of instances) on a wide variety of commodity hardware, including a laptop computer.

Knowledge Base Semantic Integration Using Crowdsourcing

The semantic web has enabled the creation of a growing number of knowledge bases (KBs), which are designed independently using different techniques. Integration of KBs has attracted much attention as different KBs usually contain overlapping and complementary information. Automatic techniques for KB integration have been improved but far from perfect. Therefore, in this paper, we study the problem of knowledge base semantic integration using crowd intelligence. There are both classes and instances in a KB, in our work, we propose a novel hybrid framework for KB semantic integration considering the semantic heterogeneity of KB class structures. We first perform semantic integration of the class structures via crowdsourcing, then apply the blocking-based instance matching approach according to the integrated class structure. For class structure (taxonomy) semantic integration, the crowd is leveraged to help identifying the semantic relationships between classes to handle the semantic heterogeneity problem. Under the conditions of both large scale KBs and limited monetary budget for crowdsourcing, we formalize the class structure (taxonomy) semantic integration problem as a Local Tree Based Query Selection (LTQS) problem. We show that the LTQS problem is NP-hard and propose two greedy-based algorithms, i.e., static query selection and adaptive query selection. Furthermore, the KBs are usually of large scales and have millions of instances, direct pairwise-based instance matching is inefficient. Therefore, we adopt the blocking-based strategy for instance matching, taking advantage of the class structure (taxonomy) integration result. The experiments on real large scale KBs verify the effectiveness and efficiency of the proposed approaches.

Large-scale hierarchical classification with rare categories and inconsistencies

Structuring data is crucial for managing massive amount of available data. Hierarchy (taxonomy) provides a natural and convenient way to organize the information. It has been extensively used in several domains, such as gene taxonomy for organizing gene sequences, international patent hierarchy for easy browsing and retrieval of patent documents, DMOZ taxonomy for web-pages categorization, and ImageNet database for indexing millions of images according to WordNet hierarchy. Given, a hierarchy containing thousands of classes (categories) and millions of instances (examples), there is an essential need to develop an efficient and automated approaches to categorize unlabeled test instances. This problem is referred to as Hierarchical Classification (HC) task. HC is an important machine learning problem that has been researched and explored extensively in the past few years (Silla Jr &amp; Freitas, 2011). The popularity of large-scale HC problem is evident from various HC competitions organized in the past few years such as LSHTC 1 , BioASQ 2 and ILSVRC 3 . HC poses several challenges due to the following reasons: (i) Data imbalance with large number of classes having very few positive examples for training (rare categories), (ii) Multi-label classification, (iii) Feature selection, (iv) Inconsistent hierarchy due to domain experts manual design, and (v) Scalability due to large number of examples, features and classes. Several approaches that address these issues individually (or multiple issues together) have been developed over the years (Gopal &amp; Yang, 2013; Babbar et al., 2013), however there are many possibilities of improving the existing methods. Specifically, we have developed the methods for handling rare categories and inconsistent hierarchy problem.