Nearest-neighbor Relationships Research Articles

Multi-target feature selection with subspace learning and manifold regularization

Existing supervised Multi-Target Feature Selection (MTFS) methods seldom consider the nearest-neighbor relationship and statistical correlation of samples underlying the output space, which leads the result of feature selection to be easily interfered by the output noise, thus making it difficult to achieve satisfactory performance. This paper proposes a novel MTFS method to preserve both the global and local target correlations. Specifically, the low-rank constraint is introduced to achieve multi-layer regression structure to better decouple the inter-input and inter-target relationships. Moreover, the local nearest-neighbor relationships and variable correlations of the sample points in the output space are also explored through adaptive graph and manifold learning, to better utilize the target correlations to improve the MTFS performance. Following the above principle, the resulting objective function and the corresponding optimization algorithm are proposed. Extensive experiments on several public datasets show that the proposed method is superior to other state-of-the-art methods.

A Probabilistic Transformation of Distance-Based Outliers

The scores of distance-based outlier detection methods are difficult to interpret, and it is challenging to determine a suitable cut-off threshold between normal and outlier data points without additional context. We describe a generic transformation of distance-based outlier scores into interpretable, probabilistic estimates. The transformation is ranking-stable and increases the contrast between normal and outlier data points. Determining distance relationships between data points is necessary to identify the nearest-neighbor relationships in the data, yet most of the computed distances are typically discarded. We show that the distances to other data points can be used to model distance probability distributions and, subsequently, use the distributions to turn distance-based outlier scores into outlier probabilities. Over a variety of tabular and image benchmark datasets, we show that the probabilistic transformation does not impact outlier ranking (ROC AUC) or detection performance (AP, F1), and increases the contrast between normal and outlier score distributions (statistical distance). The experimental findings indicate that it is possible to transform distance-based outlier scores into interpretable probabilities with increased contrast between normal and outlier samples. Our work generalizes to a wide range of distance-based outlier detection methods, and, because existing distance computations are used, it adds no significant computational overhead.

Possible new phase transition in the 3D Ising model associated with boundary percolation

In the ordered phase of the 3D Ising model, minority spin clusters are surrounded by a boundary of dual plaquettes. As the temperature is raised, these spin clusters become more numerous, and it is found that eventually their boundaries undergo a percolation transition when about 13% of spins are minority. Boundary percolation differs from the more commonly studied site and link percolation, although it is related to an unusual type of site percolation that includes next to nearest neighbor relationships. Because the Ising model can be reformulated in terms of the domain boundaries alone, there is reason to believe boundary percolation should be relevant here. A symmetry-breaking order parameter is found in the dual theory, the 3D gauge Ising model. It is seen to undergo a phase transition at a coupling close to that predicted by duality from the boundary percolation. This transition lies in the disordered phase of the gauge theory and has the nature of a spin-glass transition. Its critical exponent is seen to match the finite-size shift exponent of the percolation transition further cementing their connection. This predicts a very weak specific heat singularity with exponent . The third energy cumulant fits well to the expected non-infinite critical behavior in a manner consistent with both the predicted exponent and critical point, indicating a true thermal phase transition. Unlike random boundary percolation, the Ising boundary percolation has two different ν exponents, one associated with largest-cluster scaling and the other with finite-size transition-point shift. This suggests there may be two different correlation lengths present.

Adaptive Graph Embedded Preserving Projection Learning for Feature Extraction and Selection

Preserving projection learning has been widely used in feature extraction and selection for unsupervised image classification. Generally, some related methods constructed a graph to represent the nearest neighbor relationships of the data based on the Euclidean distances among different samples, which used 0 or 1 to predefine whether two samples are from the same class. Since a simple Euclidean distance is sensitive to noise, the predefined graph cannot produce exact correlations between the two samples. What is more, the predefined graph cannot reflect the structure of the projected data on a latent subspace when the projection matrix is learned. To solve these problems, in this article a novel adaptive graph embedded preserving projection learning (AGE_PPL) method is proposed, first combining the sparsity-based graph learning and the projection learning as an integral framework for feature extraction and feature selection. In particular, a sparse representation term with <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$l_{1}$ </tex-math></inline-formula> -norm is exploited in AGE_PPL to achieve the adaptive graph of the data to preserve the local structures among different samples while the projection matrix is learned. Meanwhile, a global-scale constraint is imposed to preserve the global structure of the data on a latent subspace. Therefore, the transformed samples will be more discriminative, allowing margins of the same class to be reduced, and margins among different classes to be enlarged. Experimental results proved the effectiveness of the proposed algorithm by obtaining competitive performances over other baseline and state-of-the-art methods. In addition, the proposed method is very flexible for feature selection and dimensionality reduction.

A study on the effects of normalized TSP features for automated algorithm selection

Classic automated algorithm selection (AS) for (combinatorial) optimization problems heavily relies on so-called instance features, i.e., numerical characteristics of the problem at hand ideally extracted with computationally low-demanding routines. For the traveling salesperson problem (TSP) a plethora of features have been suggested. Most of these features are, if at all, only normalized imprecisely raising the issue of feature values being strongly affected by the instance size. Such artifacts may have detrimental effects on algorithm selection models. We propose a normalization for two feature groups which stood out in multiple AS studies on the TSP: (a) features based on a minimum spanning tree (MST) and (b) nearest neighbor relationships of the input instance. To this end we theoretically derive minimum and maximum values for properties of MSTs and k-nearest neighbor graphs (NNG) of Euclidean graphs. We analyze the differences in feature space between normalized versions of these features and their unnormalized counterparts. Our empirical investigations on various TSP benchmark sets point out that the feature scaling succeeds in eliminating the effect of the instance size. A proof-of-concept AS-study shows promising results: models trained with normalized features tend to outperform those trained with the respective vanilla features.

Video recording and vegetation classification elucidate sheep foraging ecology in species-rich grassland.

Factors influencing grazing behavior in species‐rich grasslands have been little studied. Methodologies have mostly had a primary focus on grasslands with lower floristic diversity.We test the hypothesis that grazing behavior is influenced by both animal and plant factors and investigate the relative importance of these factors, using a novel combination of video technology and vegetation classification to analyze bite and step rates.In a semi‐natural, partially wooded grassland in northern Estonia, images of the vegetation being grazed and records of steps and bites were obtained from four video cameras, each mounted on the sternum of a sheep, during 41 animal‐hours of observation over five days. Plant species lists for the immediate field of view were compiled. Images were partnered by direct observation of the nearest‐neighbor relationships of the sheep. TWINSPAN, a standard vegetation classification technique allocating species lists to objectively defined classes by a principal components procedure, was applied to the species lists and 25 vegetation classes (15 open pasture and 10 woodland) were identified from the images.Taking bite and step rates as dependent variables, relative importance of animal factors (sheep identity), relative importance of day, and relative importance of plant factors (vegetation class) were investigated. The strongest effect on bite rates was of vegetation class. Sheep identity was less influential. When the data from woodland were excluded, sheep identity was more important than vegetation class as a source of variability in bite rate on open pasture.The original hypothesis is therefore supported, and we further propose that, at least with sheep in species‐rich open pastures, animal factors will be more important in determining grazing behavior than plant factors. We predict quantifiable within‐breed and between‐breed differences, which could be exploited to optimize conservation grazing practices and contribute to the sustainability of extensive grazing systems.

Systematic Data Analysis and Diagnostic Machine Learning Reveal Differences between Compounds with Single- and Multitarget Activity.

Small molecules with multitarget activity are capable of triggering polypharmacological effects and are of high interest in drug discovery. Compared to single-target compounds, promiscuity also affects drug distribution and pharmacodynamics and alters ADMET characteristics. Features distinguishing between compounds with single- and multitarget activity are currently only little understood. On the basis of systematic data analysis, we have assembled large sets of promiscuous compounds with activity against related or functionally distinct targets and the corresponding compounds with single-target activity. Machine learning predicted promiscuous compounds with surprisingly high accuracy. Molecular similarity analysis combined with control calculations under varying conditions revealed that accurate predictions were largely determined by structural nearest-neighbor relationships between compounds from different classes. We also found that large proportions of promiscuous compounds with activity against related or unrelated targets and corresponding single-target compounds formed analog series with distinct chemical space coverage, which further rationalized the predictions. Moreover, compounds with activity against proteins from functionally distinct classes were often active against unique targets that were not covered by other promiscuous compounds. The results of our analysis revealed that nearest-neighbor effects determined the prediction of promiscuous compounds and that preferential partitioning of compounds with single- and multitarget activity into structurally distinct analog series was responsible for such effects, hence providing a rationale for the presence of different structure-promiscuity relationships.

Fuzzy Smooth Equilibrium Method for Clustering

Clustering model plays an indispensable role in exploring data structures. To extend supervised learning to unsupervised, the maximum margin clustering model has been proposed. Maximum margin-based frameworks develop a powerful tool for supervised learning. It could yield good results by combining with some fuzzy clustering models. However, such methods characterized by high computational cost and are sensitive to the nearest neighbor relationships between data objects. Sometimes, they could lead to degenerate solutions. By reconstructing the Laplacian matrix with different similarity measurements, a new fuzzy smooth equilibrium clustering (FSEC) model is proposed. This model combines MMC with spectral clustering, but there is no need to solve the eigenvalue decomposition problem. Using the equilibrium regularization term can avoid degenerate solutions. Numerous experiments have established the effectiveness of the newly FSEC model.

Spatial structural characteristics of forests dominated by Pinus tabulaeformis Carr.

The Chinese pine (Pinus tabulaeformis Carr.) is an ecologically and economically important evergreen coniferous tree which dominates warm temperate forests throughout northern China. We established two permanent plots within the Chinese pine forest in the Jiulong Mountains, Beijing, China. To understand the structural characteristics and dynamics of these plots, we analyzed the spatial structural characteristics within nearest-neighbor relationships using the bivariate distributions of the stand spatial structural parameters: uniform angle index, W; mingling index, M; dominance index, U; and crowding index, C. Results revealed that most trees in the forest were randomly distributed. The predominant individuals and randomly arranged trees were in very dense areas and surrounded by the same species. In addition, both plots exhibited a uniform size differentiation pattern. The two plots differed mainly in the level of species mixture and dominance. The majority of reference trees in the pure Chinese pine forest (plot 1) exhibited poor species mingling and low dominance, whereas trees in the mixed Chinese pine forest (plot 2) were evenly distributed in each mingling class and most trees were of intermediate dominance. The study results are useful for optimizing forest management activities in the studied stands, promoting tree growth, regeneration and habitat diversity, and improving forest quality at a fine scale.

Tests of the sorption and olfactory "fovea" hypotheses in the mouse.

The spatial distribution of receptors within sensory epithelia (e.g., retina and skin) is often markedly nonuniform to gain efficiency in information capture and neural processing. By contrast, odors, unlike visual and tactile stimuli, have no obvious spatial dimension. What need then could there be for either nearest-neighbor relationships or nonuniform distributions of receptor cells in the olfactory epithelium (OE)? Adrian (Adrian ED. J Physiol 100: 459-473, 1942; Adrian ED. Br Med Bull 6: 330-332, 1950) provided the only widely debated answer to this question when he posited that the physical properties of odors, such as volatility and water solubility, determine a spatial pattern of stimulation across the OE that could aid odor discrimination. Unfortunately, despite its longevity, few critical tests of the "sorption hypothesis" exist. Here we test the predictions of this hypothesis by mapping mouse OE responses using the electroolfactogram (EOG) and comparing these response "maps" to computational fluid dynamics (CFD) simulations of airflow and odorant sorption patterns in the nasal cavity. CFD simulations were performed for airflow rates corresponding to quiet breathing and sniffing. Consistent with predictions of the sorption hypothesis, water-soluble odorants tended to evoke larger EOG responses in the central portion of the OE than the peripheral portion. However, sorption simulation patterns along individual nasal turbinates for particular odorants did not correlate with their EOG response gradients. Indeed, the most consistent finding was a rostral-greater to caudal-lesser response gradient for all the odorants tested that is unexplained by sorption patterns. The viability of the sorption and related olfactory "fovea" hypotheses are discussed in light of these findings.NEW & NOTEWORTHY Two classical ideas concerning olfaction's receptor-surface two-dimensional organization-the sorption and olfactory fovea hypotheses-were found wanting in this study that afforded unprecedented comparisons between electrophysiological recordings in the mouse olfactory epithelium and computational fluid dynamic simulations of nasal airflow. Alternatively, it is proposed that the olfactory receptor layouts in macrosmatic mammals may be an evolutionary contingent state devoid of the functional significance found in other sensory epithelia like the cochlea and retina.

Nearest neighbour relationships in Pinus yunnanensis var. tenuifolia forests along the Nanpan River, China

Abstract: Forest stand structural diversity can be examined at different scales. Small-scale structural changes are the basis of forest structural diversity and habitat heterogeneity, and play a key role in biodiversity conservation. Most research on forest structure has focused mainly at stand level and above, with little attention paid to fine-scale structure and correlations among different forest stand attributes. We set up four permanent plots within a secondary forest community of Pinus yunnanensis var. tenuifolia mixed forests along the Nanpan River in southern China. We analyzed their nearest-neighbor relationships using a bivariate distribution of stand spatial structural parameters (SSSP) with the aim of understanding the processes that drive structural diversity in the development of a secondary forest community. Our results revealed that communities with different disturbance histories and species compositions differed in the level of species mixture. Large, small, and medium-sized trees were well mixed within the community, both conspecific and heterospecificindividual with varying densities. All plots exhibited a uniform size differentiation pattern. Trees with different dominance levels or mixture levels were randomly distributed within the plots, and only few of these displayed clumped or regular distribution. Pearson’s correlation analysis revealed that distribution patterns may be related to species composition and diameter differentiation, though their relationship was very weak. The results of this study are relevant to optimize forest management activities in the studied stands, and promote tree growth, regeneration and habitat diversity at the fine scale.

Nearest neighbor methods for testing reflexivity

Nearest neighbor (NN) methods are widely employed for drawing inferences about spatial point patterns of two or more classes. We introduce a method for testing reflexivity in the NN structure (i.e., NN reflexivity) based on a contingency table which will be called reflexivity contingency table (RCT) henceforth. The RCT is based on the NN relationships among the data points and was used for testing niche specificity in literature, but we demonstrate that it is actually more appropriate for testing the NN reflexivity pattern. We derive the asymptotic distribution of the entries of the RCT under random labeling and introduce tests of reflexivity based on these entries. We also consider Pielou’s approach on RCT and show that it is not appropriate for completely mapped spatial data. We determine the appropriate null hypotheses and the underlying conditions/assumptions required for all tests considered. We investigate the finite sample performance of the tests in terms of empirical size and power by extensive Monte Carlo simulations and illustrate the methods on two real-life ecological data sets.

Investigation of spatio-temporal cancer clusters using residential histories in a case-control study of non-Hodgkin lymphoma in the United States.

BackgroundNon-Hodgkin lymphoma (NHL) is an enigmatic disease with few known risk factors. Spatio-temporal epidemiologic analyses have the potential to reveal patterns that may give clues to new risk factors worthy of investigation. We sought to investigate clusters of NHL through space and time based on life course residential histories.MethodsWe used residential histories from a population-based NHL case–control study of 1300 cases and 1044 controls with recruitment centers in Iowa, Detroit, Seattle, and Los Angeles, and diagnosed in 1998–2000. Novel methods for cluster detection allowing for residential mobility, called Q-statistics, were used to quantify nearest neighbor relationships through space and time over the life course to identify cancer clusters. Analyses were performed on all cases together and on two subgroups of NHL: Diffuse large B-cell lymphoma and follicular lymphoma. These more homogenous subgroups of cases might have a more common etiology that could potentially be detected in cluster analysis. Based on simulation studies designed to help account for multiple testing across space and through time, we required at least four significant cases nearby one another to declare a region a potential cluster, along with confirmatory analyses using spatial-only scanning windows (SaTScan).ResultsEvidence of a small cluster in southeastern Oakland County, MI was suggested using residences 10–18 years prior to diagnosis, and confirmed by SaTScan in a time-slice analysis 20 years prior to diagnosis, when all cases were included in the analysis. Consistent evidence of clusters was not seen in the two histologic subgroups.ConclusionsSuggestive evidence of a small space-time cluster in southeastern Oakland County, MI was detected in this NHL case–control study in the USA.

Towards a comprehensive view of the neural‐electrode tissue interface: image analysis for neural engineering applications (153.8)

Biocompatibility issues limit the longevity of neural interfacing devices designed to restore sensory‐motor function. However, a comprehensive systems‐level understanding of the device‐tissue interface is lacking. We combined immunohistochemistry, high‐resolution digital slide scanning and computational image analyses to evaluate the response associated with floating microelectrode arrays implanted into feline dorsal (sensory) or ventral (motor) lumbosacral roots (n = 16 animals; 2 implants/animal). Paraffin‐embedded sections were stained with a variety of chemical (hematoxylin/eosin, Nissl/Luxol fast blue) and antibody‐based [NF200 (axons), S100 (Schwann cells), vimentin (fibroblasts, endothelial cells, astrocytes), NeuN (neuronal nuclei), caspase‐3 (apoptotic cell death), MAC387 (monocytes/macrophages), GLUT1 (glucose transporter)] stains. Digital images were analyzed using NearCyte software. Intensity of the inflammatory reaction as a function of distance from the electrode‐tissue interface was quantified. Neural degeneration in the spinal cord proximal and distal to the site of array implant was compared to the contralateral, non‐implanted side. Neuronal density and nearest neighbor relationships were mapped and electrode performance metrics considered. This work provides a more comprehensive, quantitative profile of the neural electrode‐tissue interface. Ongoing work aims to incorporate multiplexed immunohistochemistry such that more in‐depth social interactions can be assessed in the context of the tissue architecture.Grant Funding Source: Supported by DARPA Contract N66001‐11‐C‐4171.

Testing Spatial Symmetry Using Contingency Tables Based on Nearest Neighbor Relations

We consider two types of spatial symmetry, namely, symmetry in the mixed or shared nearest neighbor (NN) structures. We use Pielou's and Dixon's symmetry tests which are defined using contingency tables based on the NN relationships between the data points. We generalize these tests to multiple classes and demonstrate that both the asymptotic and exact versions of Pielou's first type of symmetry test are extremely conservative in rejecting symmetry in the mixed NN structure and hence should be avoided or only the Monte Carlo randomized version should be used. Under RL, we derive the asymptotic distribution for Dixon's symmetry test and also observe that the usual independence test seems to be appropriate for Pielou's second type of test. Moreover, we apply variants of Fisher's exact test on the shared NN contingency table for Pielou's second test and determine the most appropriate version for our setting. We also consider pairwise and one-versus-rest type tests in post hoc analysis after a significant overall symmetry test. We investigate the asymptotic properties of the tests, prove their consistency under appropriate null hypotheses, and investigate finite sample performance of them by extensive Monte Carlo simulations. The methods are illustrated on a real-life ecological data set.

Incomplete mixing versus clathrate-like structures: A molecular view on hydrophobicity in methanol–water mixtures

The underlying molecular mechanisms of macroscopic excess properties were studied by molecular dynamics simulations for different compositions of methanol-water mixtures. Structural data (nearest neighbor relationships, clustering analysis) and dynamic data (hydrogen bond lifetimes, rotational autocorrelation, translational diffusion) were evaluated. Nearest neighbor relationships provide quantitative evidence and a pictorial description of incomplete mixing at the molecular level as a source for mixture anomalies, while a comparative study of water surrounding methyl moieties versus water in the bulk-like environment provides evidence against the hydrophobicity model of clathrate-like hydration. Furthermore, the formation or breakdown of the system-wide hydrogen bonding network at a critical threshold of approximately equimolar mixture is perceived to separate the mixture system into two hydrogen bonding regimes: hydrogen-bonded water clusters embedded in methanol for mixtures with low water content and methanol molecules within a system-wide hydrogen-bonded water network for mixtures with high water content.

Performance of cancer cluster Q-statistics for case-control residential histories

Few investigations of health event clustering have evaluated residential mobility, though causative exposures for chronic diseases such as cancer often occur long before diagnosis. Recently developed Q-statistics incorporate human mobility into disease cluster investigations by quantifying space- and time-dependent nearest neighbor relationships. Using residential histories from two cancer case-control studies, we created simulated clusters to examine Q-statistic performance. Results suggest the intersection of cases with significant clustering over their life course, Qi, with cases who are constituents of significant local clusters at given times, Qit, yielded the best performance, which improved with increasing cluster size. Upon comparison, a larger proportion of true positives were detected with Kulldorf’s spatial scan method if the time of clustering was provided. We recommend using Q-statistics to identify when and where clustering may have occurred, followed by the scan method to localize the candidate clusters. Future work should investigate the generalizability of these findings.

Surface anatomy as an entryway to clinical anatomy

Like clinicians, we begin our course with a physical exam. Students palpate their donor and paint the bony landmarks that will help them locate and paint the surface projection of structures included in a general physical exam. The exercise introduces the skeleton in relation to the underlying organs. The body painting then guides common surgical incisions. Through these incisions, students discover whether their donor had an appendectomy, an enlarged liver, and so on. This correlates surface landmarks with underlying structures and clinical applications. Surface anatomy is revisited in subsequent labs by focusing on the history, physical, and surgical procedures of actual clinical cases. Surface anatomy is further emphasized by performing simulated nerve blocks. Surface anatomy locates the insertion point for the needle and the deep structures that will be palpated with the tip of the needle. Students learn where nearby, unintended targets are located and the consequences of injuring them. Instead of anesthetic, the students inject paint. Dissection reveals the painted structures to emphasize nearest‐neighbor relationships. The faculty that subsequently teaches physical exam reports that students retain more anatomy, which allows the class to make greater progress on the clinical content. The clinical faculty's enthusiasm is further demonstrated by their active participation in our lab exercises.

Validity of spatial models of arsenic concentrations in private well water

Objective Arsenic is a pervasive contaminant in underground aquifers worldwide, yet documentation of health effects associated with low-to-moderate concentrations (<100 μg/L) has been stymied by uncertainties in assessing long-term exposure. A critical component of assessing exposure to arsenic in drinking water is the development of models for predicting arsenic concentrations in private well water in the past; however, these models are seldom validated. The objective of this paper is to validate alternative spatial models of arsenic concentrations in private well water in southeastern Michigan. Methods From 1993 to 2002, the Michigan Department of Environmental Quality analyzed arsenic concentrations in water from 6050 private wells. This dataset was used to develop several spatial models of arsenic concentrations in well water: proxy wells based on nearest-neighbor relationships, averages across geographic regions, and geostatistically derived estimates based on spatial correlation and geologic factors. Output from these models was validated using arsenic concentrations measured in 371 private wells from 2003 to 2006. Results The geostatisical model and nearest-neighbor approach outperformed the models based on geographic averages. The geostatistical model produced the highest degree of correlation using continuous data (Pearson's r=0.61; Spearman's rank ρ=0.46) while the nearest-neighbor approach produced the strongest correlation ( κ weighted=0.58) using an a priori categorization of arsenic concentrations (<5, 5–9.99, 10–19.99, ⩾20 μg/L). When the maximum contaminant level was used as a cut-off in a two-category classification (<10, ⩾10 μg/L), the nearest-neighbor approach and geostatistical model had similar values for sensitivity (0.62–0.63), specificity (0.80), negative predictive value (0.85), positive predictive value (0.53), and percent agreement (75%). Discussion This validation study reveals that geostatistical modeling and nearest-neighbor approaches are effective spatial models for predicting arsenic concentrations in private well water. Further validation analyses in other regions are necessary to indicate how widely these findings may be generalized.

Modeling Recrystallization in Aluminum Using Input from Experimental Observations

A model has been constructed for the microstructural evolution that occurs during the annealing of aluminum alloys. Geometric and crystallographic observations from two orthogonal sections through a polycrystal using automated Electron Back-Scatter Diffraction (EBSD) were used as an input to the computer simulations to create a statistically representative threedimensional model. The microstructure is generated using a voxel-based tessellation technique. Assignment of orientations to the grains is controlled to ensure that both texture and nearest neighbor relationships match the observed distributions. The microstructures thus obtained are allowed to evolve using a Monte-Carlo simulation. Anisotropic grain boundary properties are used in the simulations. Nucleation is done in accordance with experimental observations on the likelihood of occurrences in particular neighborhoods. We will present the effect of temperature on the model predictions.