Negative Instances Research Articles

Maximizing the Diversity of Ensemble Random Forests for Tree Genera Classification Using High Density LiDAR Data

Recent research into improving the effectiveness of forest inventory management using airborne LiDAR data has focused on developing advanced theories in data analytics. Furthermore, supervised learning as a predictive model for classifying tree genera (and species, where possible) has been gaining popularity in order to minimize this labor-intensive task. However, bottlenecks remain that hinder the immediate adoption of supervised learning methods. With supervised classification, training samples are required for learning the parameters that govern the performance of a classifier, yet the selection of training data is often subjective and the quality of such samples is critically important. For LiDAR scanning in forest environments, the quantification of data quality is somewhat abstract, normally referring to some metric related to the completeness of individual tree crowns; however, this is not an issue that has received much attention in the literature. Intuitively the choice of training samples having varying quality will affect classification accuracy. In this paper a Diversity Index (DI) is proposed that characterizes the diversity of data quality (Qi) among selected training samples required for constructing a classification model of tree genera. The training sample is diversified in terms of data quality as opposed to the number of samples per class. The diversified training sample allows the classifier to better learn the positive and negative instances and; therefore; has a higher classification accuracy in discriminating the “unknown” class samples from the “known” samples. Our algorithm is implemented within the Random Forests base classifiers with six derived geometric features from LiDAR data. The training sample contains three tree genera (pine; poplar; and maple) and the validation samples contains four labels (pine; poplar; maple; and “unknown”). Classification accuracy improved from 72.8%; when training samples were selected randomly (with stratified sample size); to 93.8%; when samples were selected with additional criteria; and from 88.4% to 93.8% when an ensemble method was used.

Concept Teaching in Instrumental Music Education

This article is a review of research literature on the teaching of concepts in instrumental music education. It is organized in four parts (a) the value of concept teaching in large instrumental ensembles, (b) time spent teaching concepts during rehearsals, (c) approaches to concept teaching, and (d) implications for music education. Research has indicated that there is value to teaching conceptually. Time invested in teaching concepts has not been found to detract from performance goals; on the contrary, performance may be enhanced. Musical concepts may be taught via positive and negative instances of concepts, music terms and descriptions, modeling, and music literature. More research on concept teaching in instrumental music is warranted.

Robust multiple-instance learning ensembles using random subspace instance selection

Many real-world pattern recognition problems can be modeled using multiple-instance learning (MIL), where instances are grouped into bags, and each bag is assigned a label. State-of-the-art MIL methods provide a high level of performance when strong assumptions are made regarding the underlying data distributions, and the proportion of positive to negative instances in positive bags. In this paper, a new method called Random Subspace Instance Selection (RSIS) is proposed for the robust design of MIL ensembles without any prior assumptions on the data structure and the proportion of instances in bags. First, instance selection probabilities are computed based on training data clustered in random subspaces. A pool of classifiers is then generated using the training subsets created with these selection probabilities. By using RSIS, MIL ensembles are more robust to many data distributions and noise, and are not adversely affected by the proportion of positive instances in positive bags because training instances are repeatedly selected in a probabilistic manner. Moreover, RSIS also allows the identification of positive instances on an individual basis, as required in many practical applications. Results obtained with several real-world and synthetic databases show the robustness of MIL ensembles designed with the proposed RSIS method over a range of witness rates, noisy features and data distributions compared to reference methods in the literature.

Object tracking via Dirichlet process-based appearance models

Object tracking is the process of locating objects of interest in video frames. Challenges still exist in handling appearance changes in object tracking for robotic vision. In this paper, we propose a novel Dirichlet process-based appearance model (DPAM) for tracking. By explicitly introducing a new model variable into the traditional Dirichlet process, we model the negative and positive target instances as the combination of multiple appearance models. Within each model, target instances are dynamically clustered based on their visual similarity. DPAM provides an infinite nonparametric mixture of distributions that can grow automatically with the complexity of the appearance data. In addition, prior off-line training or specifying the number of mixture components (clusters or parameters) is not required. We build a tracking system in which DPAM is applied to cluster negative and positive target samples and detect the new target location. Our experimental results on real-world videos show that our system achieves superior performance when compared with several state-of-the-art trackers.

Automated Capture and Execution of Manufacturability Rules Using Inductive Logic Programming

Capturing domain knowledge can be a time-consuming process that typically requires the collaboration of a Subject Matter Expert and a modeling expert to encode the knowledge. In a number of domains and applications, this situation is further exacerbated by the fact that the Subject Matter Expert may find it difficult to articulate the domain knowledge as a procedure or rules, but instead may find it easier to classify instance data. To facilitate this type of knowledge elicitation from Subject Matter Experts, we have developed a system that automatically generates formal and executable rules from provided labeled instance data. We do this by leveraging the techniques of Inductive Logic Programming (ILP) to generate Horn clause based rules to separate out positive and negative instance data. We illustrate our approach on a Design For Manufacturability (DFM) platform where the goal is to design products that are easy to manufacture by providing early manufacturability feedback. Specifically we show how our approach can be used to generate feature recognition rules from positive and negative instance data supplied by Subject Matter Experts. Our platform is interactive, provides visual feedback and is iterative. The feature identification rules generated can be inspected, manually refined and vetted.

Werewolves in the Immunitary Paradigm

I. IntroductionThe biopolitical perspective initiated by Michel Foucault highlights the limitations of traditional categories of political philosophy to explicate power relations. As Foucault notes, power things, it induces pleasure, forms knowledge, produces discourses. It needs to be considered as a productive network which runs through the whole social body, much more than as a negative instance whose function is repression.1 Power is a vast technology that crosses all social relations; it can produce both subjects and subjectivities.Based on this point, we will consider monstrosity as a political category2 that allows us to conceive the way in which biopower produces subjectivities. Monstrosity can be understood in two ways: in the negative sense it can be taken as subjectivities that are the negative converse of the useful and docile3 bodies. Consequently, monstrosity becomes an objective of biopower that aims to neutral ize and eliminate it. However, monstrosity can also be taken in a positive sense as subjectivities that resist appropriation by biopower and create in common.4The category of monstrosity reveals a semantic ambiguity as it consists of an inherent tension between the negative and positive meanings. These are related by the exercise of power in a specific direction. For the negative orientation, we can use the expression over monstrosity-stressing over-and of monstrosity-stressing of-for the positive direction.In this article we are going to focus on politics over monstrosity, that is, on the subjectivities that are constituted as monsters by power because they cannot be integrated into (captured by) the logic of normalization, whose aim is homogenization, or subsumed under classifications in binary oppositions-such as, male/female and capitalist/worker.5 These subjectivities can be seen as a way of life, a negative way of life, which can be observed in the expressions life unworthy of being lived outlined by Giorgio Agamben or the life without grievability according to Judith Butler,6 for example.We would like to refer to an old metaphor which allows us to understand the inherent relation between State and monstrosity, when monstrosity is introduced in the State. That is, the werewolf metaphor, which summarizes the meaning of a sentence used by Thomas Hobbes:7 homo homini lupus, man is a wolf to man. We will thereby focus on the monstrosity that the human being represents, and not on the Leviathan, which is usually recognized as a political monster.8Even though is hostile to the use of metaphors in sciences,9 he uses them extensively. As Robert Stillman argues, Hobbes is fond of metaphors of the monstrous, and his employment of them, especially in crucial accounts of his own vocational ambitions, is recurrent and revealing.10 However, this should not be seen as a contradiction in his thought, but rather as a rhetorical strategy.11 In this respect, we note that only refers to the Leviathan three times in the voluminous book that takes this name. In a similar way, although the sentence homo homini lupus is used only once by Hobbes, this idea imbued his anthropological conception.Hobbes's thought is relevant from a biopolitical perspective as he was the first philosopher to demonstrate the direct relation between and politics.12 Politics is perceived as a protection against violent death, which is the cause of fear among men. It is out of fear of being murdered, which can be interpreted as political existentialism, that men create the State. In this sense, politics is understood as conservatio vitae, a preservation of against the constant threat of death in the state of nature.In this paper, we do not intend to analyze Hobbes's political theory but rather to look at the werewolf figure, which allows us to understand monstrosity from a biopolitical paradigm. This figure has traditionally been conceived as a mix of species, a combination of human and beast and, as Jeffrey Cohen indicates, the werewolf can express the radical difference between men and beasts or a hesitation during which, what is supposed to be contrastive remains coexistent, indifferent. …

Transient elastography and APRI score: looking at false positives and false negatives. Diagnostic performance and association to fibrosis staging in chronic hepatitis C

Although long regarded as the gold standard for liver fibrosis staging in chronic hepatitis C (CHC), liver biopsy (LB) implies both the risk of an invasive procedure and significant variability. The aim of this study was to evaluate the diagnostic performance for transient elastography (TE) and aspartate aminotransferase to platelet index (APRI) used alone and in combination compared to liver biopsy and to analyze false positive/negative results. Patients with CHC, and no previous clinical diagnosis of cirrhosis were enrolled to undergo liver biopsy, TE and APRI. A total of 182 adult patients with a median age of 55 years and median body mass index of 26.71 kg/m2 were analyzed. On LB, 56% of patients had significant levels of fibrosis (METAVIR F≥2) and 28% had advanced fibrosis (F3/F4). The strongest performance for both tests was observed for exclusion of advanced fibrosis with good negative predictive values (89 and 86%, respectively). Low necroinflammatory activity on LB was associated with false negative TE. False positives were associated with NASH and smaller LB fragments. Correlation between APRI and Fibroscan for F≥2 was 100% and 84% for F≥3 and remained high in both false negative and false positive instances, correctly identifying F<2 in 71% of cases and F<3 in 78% (and potentially foregoing up to 84% of LB). We concluded that low individual performance indicators could be attributable to limitations of LB. Poorer differentiation of lower levels of fibrosis is a known issue for LB and remains so for noninvasive tests. Good predictability is possible, however, for advanced fibrosis.

Utility of Responsiveness Theory for Classifying Supportive Behaviors to Enhance Smokeless Tobacco Cessation.

Although social support is correlated with successful tobacco cessation, interventions designed to optimize social support have shown mixed results. Understanding the process of providing social support for tobacco cessation may suggest new approaches to intervention. Responsiveness theory provides a new framework for classifying supportive behaviors in the context of tobacco cessation. It proposes three main components to sustaining relationship quality when providing support to an intimate partner: showing respect, showing understanding, and showing caring. Interviews were conducted with 35 women whose husbands or domestic partners had quit smokeless tobacco and were analyzed within a responsiveness theory framework: Positive and negative instances of the three supportive components were expressed in terms of beliefs and attitudes, interactions with the chewer, and behaviors outside of the interaction context. Positive activities included respecting the chewer's decision on whether, when, and how to quit; perspective-taking and other efforts to understand his subjective experience; and expressing warmth and affection toward the chewer. Particularly problematic for the women were the challenges of respecting the chewer's autonomy (ie, negative behaviors such as nagging him to quit or monitoring his adherence to his cessation goal) and lack of understanding the nature of addiction. The findings help to confirm the potential utility of responsiveness theory for elucidating the breadth of both positive and negative forms of partner support that may be useful to guide social support interventions for tobacco cessation. The study provides a categorization system for positive and negative social support during smokeless tobacco cessation, based on responsiveness theory and interviews with 35 partners of smokeless users.

The classification of imbalanced large data sets based on MapReduce and ensemble of ELM classifiers

Aiming at effectively classifying imbalanced large data sets with two classes, this paper proposed a novel algorithm, which consists of four stages: (1) alternately over-sample p times between positive class instances and negative class instances; (2) construct l balanced data subsets based on the generated positive class instances; (3) train l component classifiers with extreme learning machine (ELM) algorithm on the constructed l balanced data subsets; (4) integrate the l ELM classifiers with simple voting approach. Specifically, in first stage, we firstly calculate the center of positive class instances, and then sample instance points along the line between the center and each positive class instance. Next, for each instance point in the new positive class, we firstly find its k nearest neighbors in negative class instances with MapRedcue, and then sample instance points along the line between the instance and its k nearest negative neighbors. The process of over-sampling is repeated p times. In the second stage, we sample instances l times from the negative class with the same size as the generated positive class instances. Each round of sampling, we put positive class and negative class instances together thus obtain l balanced data subsets. The experimental results show that the proposed algorithm can obtain promising speed-up and scalability, and also outperforms three other ensemble algorithms in G-mean.

Superfluous Negation in Modern Hebrew and Its Origins

In this article, we survey a variety of constructions in contemporary Modern Hebrew that include seemingly superfluous instances of negation. These include free relatives, exclamative rhetorical questions, clausal complements of ‘until,’ ‘without,’ and ‘before,’ clausal complements of ‘fear’-type verbs, after negated ‘surprise,’ and the complement of ‘almost’ (a construction by now obsolete). We identify possible sources for these constructions in pre-modern varieties of Hebrew. When an earlier source cannot be found, we examine earliest attestations of the constructions in modern-era corpora and consider the role of contact (primarily with Yiddish and Slavic) in their development.

Physical and biological pretreatment quality assurance of the head and neck cancer plan with the volumetric modulated arc therapy

The aim of this work is to demonstrate both the physical and the biological quality assurance (QA) aspects as pretreatment QA of the head and neck (H&N) cancer plan for the volumetric modulated arc therapy (VMAT). Ten H&N plans were studied. The COMPASS® dosimetry analysis system and the tumor control probability (TCP) and the normal tissue complication probability (NTCP) calculation free program were used as the respective measurement and calculation tools. The reliability of these tools was verified by a benchmark study in accordance with the TG-166 report. For the physical component of QA, the gamma passing rates and the false negative cases between the calculated and the measured data were evaluated. The biological component of QA was performed based on the equivalent uniform dose (EUD), TCP and NTCP values. The evaluation was performed for the planning target volumes (PTVs) and the organs at risks (OARs), including the eyes, the lens, the parotid glands, the esophagus, the spinal cord, and the brainstem. All cases had gamma passing rates above 95% at an acceptance tolerance level with the 3%/3 mm criteria. In addition, the false negative instances were presented for the PTVs and OARs. The gamma passing rates exhibited a weak correlation with false negative cases. For the biological QA, the physical dose errors affect the EUD and the TCP for the PTVs, but no linear correlation existed between them. The EUD and NTCP for the OARs were shown the random differences that could not be attributed to the dose errors from the physical QA. The differences in the EUD and NTCP between the calculated and the measured results were mainly demonstrated for the parotid glands. This study describes the importance and the necessity of improved QA to accompany both the physical and the biological aspects for accurate radiation treatment.

Lead curve detection in drawings with complex cross-points

Lead curve detection in design drawings is a critical problem in a wide range of applications ranging from checking similar drawings in patent granting to constructing hyperlinks between image and text description in digitalization. The difficulty of the problem are two folds: unknown end point of lead curve and complex crossings. However, most previous curve detection algorithms are usually applied in simple or no crossing situations. We make four contributions in solving the problem: (1) we transform the problem into a new problem that finds an optimal path with the best score in the cross-point graph. We introduce the “cross-point graph” representation which captures the topology of cross-point connectedness. Based on the original drawing and the corresponding cross-point graph, we introduce the coupling concept “curve-path”, which correlates the curve in the original drawing with the corresponding path in the cross-point graph. (2) we design a set of joint feature representations for curve-path which describes different characteristics of a curve and its corresponding path. (3) we define a task specific loss function for our customized structured SVM. We propose a mixed negative instance sampling strategy to learn the weights of different joint feature representations. We prune the search space effectively for fast lead curve detection. (4) we build a software to efficiently facilitate manual lead curve labeling. We release the patent drawing dataset with groundtruth to public for lead curve detection research. The extensive experimental results prove the effectiveness of our methods.

Automated Detection of HONcode Website Conformity Compared to Manual Detection: An Evaluation.

BackgroundTo earn HONcode certification, a website must conform to the 8 principles of the HONcode of Conduct In the current manual process of certification, a HONcode expert assesses the candidate website using precise guidelines for each principle. In the scope of the European project KHRESMOI, the Health on the Net (HON) Foundation has developed an automated system to assist in detecting a website’s HONcode conformity. Automated assistance in conducting HONcode reviews can expedite the current time-consuming tasks of HONcode certification and ongoing surveillance. Additionally, an automated tool used as a plugin to a general search engine might help to detect health websites that respect HONcode principles but have not yet been certified.ObjectiveThe goal of this study was to determine whether the automated system is capable of performing as good as human experts for the task of identifying HONcode principles on health websites.MethodsUsing manual evaluation by HONcode senior experts as a baseline, this study compared the capability of the automated HONcode detection system to that of the HONcode senior experts. A set of 27 health-related websites were manually assessed for compliance to each of the 8 HONcode principles by senior HONcode experts. The same set of websites were processed by the automated system for HONcode compliance detection based on supervised machine learning. The results obtained by these two methods were then compared.ResultsFor the privacy criterion, the automated system obtained the same results as the human expert for 17 of 27 sites (14 true positives and 3 true negatives) without noise (0 false positives). The remaining 10 false negative instances for the privacy criterion represented tolerable behavior because it is important that all automatically detected principle conformities are accurate (ie, specificity [100%] is preferred over sensitivity [58%] for the privacy criterion). In addition, the automated system had precision of at least 75%, with a recall of more than 50% for contact details (100% precision, 69% recall), authority (85% precision, 52% recall), and reference (75% precision, 56% recall). The results also revealed issues for some criteria such as date. Changing the “document” definition (ie, using the sentence instead of whole document as a unit of classification) within the automated system resolved some but not all of them.ConclusionsStudy results indicate concordance between automated and expert manual compliance detection for authority, privacy, reference, and contact details. Results also indicate that using the same general parameters for automated detection of each criterion produces suboptimal results. Future work to configure optimal system parameters for each HONcode principle would improve results. The potential utility of integrating automated detection of HONcode conformity into future search engines is also discussed.

Recognition improvement through the optimisation of learning instances

Image recognition based on machine learning has been widely utilised in the computer vision field. In the image recognition process, quite a few positive and negative instances are needed for effective machine learning. However, some invalid instances selected from the instance candidates, particularly for negative instances, will result in reduced image recognition accuracy and wasted resources. When making the instance selections for machine learning, if a large number of negative instances are selected that exhibit a semantic distance too close to the positive instances, the recognition accuracy will be considerably reduced. The selection of valid negative instances from the instance candidates has become a new challenge in the field of computer vision. In this paper a new method containing several different algorithms is presented, that more effectively selects valid negative instances. In this innovative process, a Wordnet subtree, semantic distance based on feature descriptors and an improved K‐nearest neighbour model were all utilised. Experiments were implemented using a large database to determine the effectiveness of this method. The results have demonstrated that this method can significantly improve the recognition results.

Semi-Supervised Outlier Detection with Only Positive and Unlabeled Data Based on Fuzzy Clustering

The task of semi-supervised outlier detection is to find the instances that are exceptional from other data, using some labeled examples. In many applications such as fraud detection and intrusion detection, this issue becomes more important. Most existing techniques are unsupervised. On the other hand, semi-supervised approaches use both negative and positive instances to detect outliers. However, in many real world applications, very few positive labeled examples are available. This paper proposes an innovative approach to address this problem. The proposed method works as follows. First, some reliable negative instances are extracted by a kNN-based algorithm. Afterwards, fuzzy clustering using both negative and positive examples is utilized to detect outliers. Experimental results on real data sets demonstrate that the proposed approach outperforms the previous unsupervised state-of-the-art methods in detecting outliers.

Not a light negation

In languages like Dutch and German, certain instances of negation, such as negated indefinites, are ruled out in out-of-the-blue contexts, i.e. with the unmarked reading where negation outscopes an unfocused indefinite (*Er kann nicht eine Fremdsprache ('He speaks not a foreign language')). Such instances are, however, fine in NPI licensing contexts. Schwarz (2004) and Bhatt &amp; Schwarz (2006) argue that in such cases the negative marker is not a plain negation, but rather a homophonous marker of light negation, which they take to be an NPI. In this paper I argue, though, that this phenomenon can be explained by adopting standard pragmatic assumptions. In short, I argue that such instances of negation are bad due to the existence of some alternative expression that only conveys the unmarked reading (in this example an expression containing a negative indefinite: er kann keine Fremdsparache (‘he speaks no foreign language’)). Uttering a less simple construction, such as the example above with the negated indefinite, will give rise to an implicature that states that the speaker does not intend such utterances with their unmarked readings, since the unmarked reading could have been conveyed in a simpler manner. This implicature, then, is suppressed in NPI licensing contexts.

Negative samples reduction in cross-company software defects prediction

ContextSoftware defect prediction has been widely studied based on various machine-learning algorithms. Previous studies usually focus on within-company defects prediction (WCDP), but lack of training data in the early stages of software testing limits the efficiency of WCDP in practice. Thus, recent research has largely examined the cross-company defects prediction (CCDP) as an alternative solution. ObjectiveHowever, the gap of different distributions between cross-company (CC) data and within-company (WC) data usually makes it difficult to build a high-quality CCDP model. In this paper, a novel algorithm named Double Transfer Boosting (DTB) is introduced to narrow this gap and improve the performance of CCDP by reducing negative samples in CC data. MethodThe proposed DTB model integrates two levels of data transfer: first, the data gravitation method reshapes the whole distribution of CC data to fit WC data. Second, the transfer boosting method employs a small ratio of labeled WC data to eliminate negative instances in CC data. ResultsThe empirical evaluation was conducted based on 15 publicly available datasets. CCDP experiment results indicated that the proposed model achieved better overall performance than compared CCDP models. DTB was also compared to WCDP in two different situations. Statistical analysis suggested that DTB performed significantly better than WCDP models trained by limited samples and produced comparable results to WCDP with sufficient training data. ConclusionsDTB reforms the distribution of CC data from different levels to improve the performance of CCDP, and experimental results and analysis demonstrate that it could be an effective model for early software defects detection.

Lift: Multi-Label Learning with Label-Specific Features.

Multi-label learning deals with the problem where each example is represented by a single instance (feature vector) while associated with a set of class labels. Existing approaches learn from multi-label data by manipulating with identical feature set, i.e. the very instance representation of each example is employed in the discrimination processes of all class labels. However, this popular strategy might be suboptimal as each label is supposed to possess specific characteristics of its own. In this paper, another strategy to learn from multi-label data is studied, where label-specific features are exploited to benefit the discrimination of different class labels. Accordingly, an intuitive yet effective algorithm named LIFT, i.e. multi-label learning with Label specific Features, is proposed. LIFT firstly constructs features specific to each label by conducting clustering analysis on its positive and negative instances, and then performs training and testing by querying the clustering results. Comprehensive experiments on a total of 17 benchmark data sets clearly validate the superiority of LIFT against other well-established multi-label learning algorithms as well as the effectiveness of label-specific features.

Predicting Treatment Relations with Semantic Patterns over Biomedical Knowledge Graphs.

Identifying new potential treatment options (say, medications and procedures) for known medical conditions that cause human disease burden is a central task of biomedical research. Since all candidate drugs cannot be tested with animal and clinical trials, in vitro approaches are first attempted to identify promising candidates. Even before this step, due to recent advances, in silico or computational approaches are also being employed to identify viable treatment options. Generally, natural language processing (NLP) and machine learning are used to predict specific relations between any given pair of entities using the distant supervision approach. In this paper, we report preliminary results on predicting treatment relations between biomedical entities purely based on semantic patterns over biomedical knowledge graphs. As such, we refrain from explicitly using NLP, although the knowledge graphs themselves may be built from NLP extractions. Our intuition is fairly straightforward - entities that participate in a treatment relation may be connected using similar path patterns in biomedical knowledge graphs extracted from scientific literature. Using a dataset of treatment relation instances derived from the well known Unified Medical Language System (UMLS), we verify our intuition by employing graph path patterns from a well known knowledge graph as features in machine learned models. We achieve a high recall (92 %) but precision, however, decreases from 95% to an acceptable 71% as we go from uniform class distribution to a ten fold increase in negative instances. We also demonstrate models trained with patterns of length ≤ 3 result in statistically significant gains in F-score over those trained with patterns of length ≤ 2. Our results show the potential of exploiting knowledge graphs for relation extraction and we believe this is the first effort to employ graph patterns as features for identifying biomedical relations.

MapReduce Based Personalized Locality Sensitive Hashing for Similarity Joins on Large Scale Data.

Locality Sensitive Hashing (LSH) has been proposed as an efficient technique for similarity joins for high dimensional data. The efficiency and approximation rate of LSH depend on the number of generated false positive instances and false negative instances. In many domains, reducing the number of false positives is crucial. Furthermore, in some application scenarios, balancing false positives and false negatives is favored. To address these problems, in this paper we propose Personalized Locality Sensitive Hashing (PLSH), where a new banding scheme is embedded to tailor the number of false positives, false negatives, and the sum of both. PLSH is implemented in parallel using MapReduce framework to deal with similarity joins on large scale data. Experimental studies on real and simulated data verify the efficiency and effectiveness of our proposed PLSH technique, compared with state-of-the-art methods.