Sparse Neighborhoods Research Articles

Granular correlation-based label-specific feature augmentation for multi-label classification

Multi-label classification is an extension of single-label classification with generations of multi-output for unseen instances. Label correlation is an essential component in constructing multi-label classifiers. How to optimize the representation of label correlation while preserving the semantics of label-specific remains an uncertain issue. Instead of estimating label correlation by a holistic feature representation, we present an augmented label correlation model by generating multi-granularity label-specific features. Firstly, we devise a mixture distance measure to characterize the closeness of an instance by weighing the Pearson correlation coefficient with cosine similarity. Secondly, we explore the local label-specific relative discrimination by leveraging from both the instance-level and class-level correlation distribution within k nearest neighborhood. Finally, we conduct an information fusion strategy to comprehensively integrate the positive and the negative tendencies at the neighborhood level. Instances with salient positive tendency and compact neighborhood structure receive larger values while receiving smaller values with salient negative tendency and sparse neighborhood structure. With the concatenation of original features and augmented features, we examine the classification performance of the proposed granule correlation-based feature augmentation (GOFA) on well-established second-order multi-label classification methods. Extensive comparisons on thirteen benchmarks demonstrate the statistical superiority of GOFA over state-of-the-art multi-label classifications.

Wayward associations: When and why people think of similar-sounding words

Words with similar meanings sometimes sound similar, which carries both risks, such as confusion, and rewards, such as ease of comprehension. It has been argued that languages evolve to balance these competing pressures, so words more often overlap in both form and meaning when they are less likely to be confused. By measuring the phonological similarity of responses to English cues in a word association megastudy, we provide evidence of a tendency to activate similar-sounding words in response to words that reside in sparse semantic neighbourhoods and in response to words for abstract concepts. Crucially, we provide evidence that the availability of similar-sounding associates helps people retrieve and represent the meanings of words from sparse neighbourhoods and words for abstract concepts, as measured by reaction time in semantic decisions and by accuracy in recognition memory. We propose that phonological connections compensate for weak semantic connections when representing word meanings, which we discuss in terms of multiplex networks, models of word-meaning access, and theories of language evolution.

Overt Word Reading and Visual Object Naming in Adults with Dyslexia: Electroencephalography Study in Transparent Orthography.

The study aimed to investigate overt reading and naming processes in adult people with dyslexia (PDs) in shallow (transparent) language orthography. The results of adult PDs are compared with adult healthy controls HCs. Comparisons are made in three phases: pre-lexical (150-260 ms), lexical (280-700 ms), and post-lexical stage of processing (750-1000 ms) time window. Twelve PDs and HCs performed overt reading and naming tasks under EEG recording. The word reading and naming task consisted of sparse neighborhoods with closed phonemic onset (words/objects sharing the same onset). For the analysis of the mean ERP amplitude for pre-lexical, lexical, and post-lexical time window, a mixed design ANOVA was performed with the right (F4, FC2, FC6, C4, T8, CP2, CP6, P4) and left (F3, FC5, FC1, T7, C3, CP5, CP1, P7, P3) electrode sites, within-subject factors and group (PD vs. HC) as between-subject factor. Behavioral response latency results revealed significantly prolonged reading latency between HCs and PDs, while no difference was detected in naming response latency. ERP differences were found between PDs and HCs in the right hemisphere's pre-lexical time window (160-200 ms) for word reading aloud. For visual object naming aloud, ERP differences were found between PDs and HCs in the right hemisphere's post-lexical time window (900-1000 ms). The present study demonstrated different distributions of the electric field at the scalp in specific time windows between two groups in the right hemisphere in both word reading and visual object naming aloud, suggesting alternative processing strategies in adult PDs. These results indirectly support the view that adult PDs in shallow language orthography probably rely on the grapho-phonological route during overt word reading and have difficulties with phoneme and word retrieval during overt visual object naming in adulthood.

Contrastive Multiview Attribute Graph Clustering With Adaptive Encoders.

Multiview attribute graph clustering aims to cluster nodes into disjoint categories by taking advantage of the multiview topological structures and the node attribute values. However, the existing works fail to explicitly discover the inherent relationships in multiview topological graph matrices while considering different properties between the graphs. Besides, they cannot well handle the sparse structure of some graphs in the learning procedure of graph embeddings. Therefore, in this article, we propose a novel contrastive multiview attribute graph clustering (CMAGC) with adaptive encoders method. Within this framework, the adaptive encoders concerning different properties of distinct topological graphs are chosen to integrate multiview attribute graph information by checking whether there exists high-order neighbor information or not. Meanwhile, the number of layers of the GCN encoders is selected according to the prior knowledge related to the characteristics of different topological graphs. In particular, the feature-level and cluster-level contrastive learning are conducted on the multiview soft assignment representations, where the union of the first-order neighbors from the corresponding graph pairs is regarded as the positive pairs for data augmentation and the sparse neighbor information problem in some graphs can be well dealt with. To the best of our knowledge, it is the first time to explicitly deal with the inherent relationships from the interview and intraview perspectives. Extensive experiments are conducted on several datasets to verify the superiority of the proposed CMAGC method compared with the state-of-the-art methods.

An efficient model for predicting human diseases through miRNA based on multiple-types of contrastive learning.

Multiple studies have demonstrated that microRNA (miRNA) can be deeply involved in the regulatory mechanism of human microbiota, thereby inducing disease. Developing effective methods to infer potential associations between microRNAs (miRNAs) and diseases can aid early diagnosis and treatment. Recent methods utilize machine learning or deep learning to predict miRNA-disease associations (MDAs), achieving state-of-the-art performance. However, the problem of sparse neighborhoods of nodes due to lack of data has not been well solved. To this end, we propose a new model named MTCL-MDA, which integrates multiple-types of contrastive learning strategies into a graph collaborative filtering model to predict potential MDAs. The model adopts a contrastive learning strategy based on topology, which alleviates the damage to model performance caused by sparse neighborhoods. In addition, the model also adopts a semantic-based contrastive learning strategy, which not only reduces the impact of noise introduced by topology-based contrastive learning, but also enhances the semantic information of nodes. Experimental results show that our model outperforms existing models on all evaluation metrics. Case analysis shows that our model can more accurately identify potential MDA, which is of great significance for the screening and diagnosis of real-life diseases. Our data and code are publicly available at: https://github.com/Lqingquan/MTCL-MDA.

A novel outlier detecting algorithm based on the outlier turning points

Outlier detection is one of the hot research in data mining, and has been applied to various fields such as network anomaly detection, image abnormal analysis, etc. In recent years, many outlier detecting algorithms have been proposed. However, these outlier detecting algorithms are hard to effectively detect global outliers, local outliers and outlier clusters at the same time. In this paper, we propose a novel outlier detecting algorithm based on the following ideas: (1) the density distribution should not be changed dramatically on local area; (2) the ratio of the number of k nearest neighbors and the number of reverse k nearest neighbors should not be very big. Based on above ideas, the proposed algorithm aims to find outlier turning points, then regards all outlier turning points and its sparse neighbors as outliers. Futhermore, the proposed algorithm use natural neighbors to obtain the neighborhood parameter k adaptively. The formal analysis and extensive experiments demonstrate that this technique can detect global outliers, local outliers and outlier clusters without neighborhood parameter k.

Complete feature learning and consistent relation modeling for few-shot knowledge graph completion

Few-shot knowledge graph completion focuses on predicting unseen facts of long-tail relations in knowledge graphs with only few reference sets. The key challenge for tackling this task is how to represent the complete entity features under low data regime conditions and further build the relation scoring function of the triplet for prediction. However, existing works mainly focus on aggregating entity representations and seriously ignore the process of consistent relation modeling, resulting in unsatisfactory performance on sparse neighbors and complex relations modeling. To address the issues, this paper designs a two-branch feature extractor to capture complementary and complete representation of entities for differentiating the few examples, where each branch focuses on diverse aspect of the entity features. Furthermore, we apply a diversity loss based on the minimization of cosine similarity is applied between the two-branch feature extractors to encourage the two-branch to learn complementary features. Conditioned on the entity features, we further incorporate the structural relation representation into the semantic relation learning to keep the consistent with triplet scoring function, and consider the consistency issue of various structural relation modeling between training and test generalization. Empirical results on two public benchmark datasets NELL-One and Wiki-One demonstrate that our approach outperforms the state-of-the-art results, with relative improvements on Hits@10 for 1-shot of 4.8% and 4.4%, respectively, and achieves new state-of-the-art results. Additionally, Extensive experiments also show proficiency in dealing with complex relations and sparse neighbors.

Modeling and design of heterogeneous hierarchical bioinspired spider web structures using deep learning and additive manufacturing

Spider webs are incredible biological structures, comprising thin but strong silk filament and arranged into complex hierarchical architectures with striking mechanical properties (e.g., lightweight but high strength, achieving diverse mechanical responses). While simple 2D orb webs can easily be mimicked, the modeling and synthesis of 3D-based web structures remain challenging, partly due to the rich set of design features. Here, we provide a detailed analysis of the heterogeneous graph structures of spider webs and use deep learning as a way to model and then synthesize artificial, bioinspired 3D web structures. The generative models are conditioned based on key geometric parameters (including average edge length, number of nodes, average node degree, and others). To identify graph construction principles, we use inductive representation sampling of large experimentally determined spider web graphs, to yield a dataset that is used to train three conditional generative models: 1) an analog diffusion model inspired by nonequilibrium thermodynamics, with sparse neighbor representation; 2) a discrete diffusion model with full neighbor representation; and 3) an autoregressive transformer architecture with full neighbor representation. All three models are scalable, produce complex, de novo bioinspired spider web mimics, and successfully construct graphs that meet the design objectives. We further propose an algorithm that assembles web samples produced by the generative models into larger-scale structures based on a series of geometric design targets, including helical and parametric shapes, mimicking, and extending natural design principles toward integration with diverging engineering objectives. Several webs are manufactured using 3D printing and tested to assess mechanical properties.

Trajectory-Free Motion Planning of an Unmanned Surface Vehicle Based on MPC and Sparse Neighborhood Graph

Trajectory-Free Motion Planning of an Unmanned Surface Vehicle Based on MPC and Sparse Neighborhood Graph

Diffusion of Phonetic Updates within Phonological Neighborhoods

Phonological neighborhood density is known to influence lexical access, speech production and perception processes. Lexical competition is considered the central concept from which the neighborhood effect emanates: highly competitive neighborhoods are characterized by large degrees of phonemic co-activation, which can delay speech recognition and facilitate speech production. The study investigates phonetic learning in English as a foreign language in relation to phonological neighborhood density and onset density to see if dense or sparse neighborhoods are more conducive to the incorporation of novel phonetic detail. Also, the effect of voice-contrasted minimal pairs is explored. The results indicate that sparser neighborhoods with weaker lexical competition provide the most optimal phonological environment for phonetic learning. Moreover, novel phonetic details are incorporated faster in neighborhoods without minimal pairs. The results indicate that lexical competition plays a role in the dissemination of phonetic updates in the lexicon of foreign language learners.

Orthographic precision for word naming in skilled readers

ABSTRACT Perfetti (2007) proposed that the quality of lexical representations affects reading. We investigated the role of lexical quality in decoding. Eighty-four adults read aloud words and pseudowords with dense and sparse neighbourhoods in a masked form-priming experiment. Individual-difference measures of language and cognitive processes were collected and entered into a principal component analysis (PCA). Compared to a non-overlapping control prime, we observed greater facilitatory form-priming for word targets with sparse neighbourhoods than those with dense neighbourhoods. A PCA component related to orthographic precision affected form-priming: people with low orthographic precision showed greater facilitation for words with sparse neighbourhoods, primed by pseudowords, than those with dense neighbourhoods. People with high orthographic precision demonstrated the converse, only when primed by words. For pseudoword reading, word primes facilitated more than pseudoword primes in people with low orthographic precision. People with high orthographic precision showed the opposite pattern.

Semi-supervised sparse neighbor constrained co-clustering with dissimilarity and similarity regularization

Nonnegative matrix factorization (NMF) is a very effective method for high dimensional data analysis, which has been widely used in computer vision. However, the conventional NMF is unsupervised, and thus, it cannot utilize the label information. To this end, semi-supervised NMF is proposed, which performs the NMF with the guidance of the supervisory information. However, semi-supervised NMF fails to make full use of label information, which limits the performance of clustering using low dimensional representation, and samples and features cannot be clustered in a mutually reinforcing manner. To solve the above shortcomings, we propose a semi-supervised sparse neighbor constrained co-clustering with dissimilarity and similarity regularization model (SSCCDS). First, co-clustering is introduced to cluster samples and features simultaneously. Secondly, the model imposes similarity and dissimilarity regularization constraints on samples by low-dimensional representations. Specifically, similarity and dissimilar regularization constraints are imposed on labeled samples, and similarity regularization constraints are imposed on unlabeled data. Thirdly, the model proposes sparse neighbor constraints for feature consistent learning. Then, a multiplicative alternating scheme is proposed for objective optimization. A large number of experiments on different data sets show that SSCCDS has good clustering ability.

Colouring graphs with sparse neighbourhoods: Bounds and applications

Let G be a graph with chromatic number χ, maximum degree Δ and clique number ω. Reed's conjecture states that χ≤⌈(1−ε)(Δ+1)+εω⌉ for all ε≤1/2. It was shown by King and Reed that, provided Δ is large enough, the conjecture holds for ε≤1/130,000. In this article, we show that the same statement holds for ε≤1/26, thus making a significant step towards Reed's conjecture. We derive this result from a general technique to bound the chromatic number of a graph where no vertex has many edges in its neighbourhood. Our improvements to this method also lead to improved bounds on the strong chromatic index of general graphs. We prove that χs′(G)≤1.835Δ(G)2 provided Δ(G) is large enough.

The company a word keeps: The role of neighbourhood density in verbal short-term memory.

Psycholinguistic information plays an important role in verbal short-term memory (vSTM). One such linguistic feature is neighbourhood density (ND)-the number of words that can be derived from a given word by changing a single phoneme or single letter-with vSTM performance typically better when word sequences are from dense rather than sparse neighbourhoods. This effect has been attributed to higher levels of supportive activation among dense neighbourhood words. Generally, it has been assumed that lexical variables influence item memory but not order memory, and we show that the typical vSTM advantage for dense neighbourhood words in serial recall is eliminated when using serial recognition. However, we also show that the usual effect of ND is reversed-for both serial recall and serial recognition-when using a subset of those same words. The findings call into question the way in which ND has been incorporated into accounts of vSTM that invoke mutual support from long-term representations on either encoding or retrieval.

Crowding, climate, and the case for social distancing among trees.

In an emerging era of megadisturbance, bolstering forest resilience to wildfire, insects, and drought has become a central objective in many western forests. Climate has received considerable attention as a driver of these disturbances, but few studies have examined the complexities of climate-vegetation-disturbance interactions. Current strategies for creating resilient forests often rely on retrospective approaches, seeking to impart resilience by restoring historical conditions to contemporary landscapes, but historical conditions are becoming increasingly unattainable amidst modern bioclimatic conditions. What becomes an appropriate benchmark for resilience when we have novel forests, rapidly changing climate, and unprecedented disturbance regimes? We combined two longitudinal datasets-each representing some of the most comprehensive spatially explicit, annual tree mortality data in existence-in a post-hoc factorial design to examine the nonlinear relationships between fire, climate, forest spatial structure, and bark beetles. We found that while prefire drought elevated mortality risk, advantageous local neighborhoods could offset these effects. Surprisingly, mortality risk (Pm ) was higher in crowded local neighborhoods that burned in wet years (Pm =42%) compared with sparse neighborhoods that burned during drought (Pm =30%). Risk of beetle attack was also increased by drought, but lower conspecific crowding impeded the otherwise positive interaction between fire and beetle attack. Antecedent fire increased drought-related mortality over short timespans (<7 years) but reduced mortality over longer intervals. These results clarify interacting disturbance dynamics and provide a mechanistic underpinning for forest restoration strategies. Importantly, they demonstrate the potential for managed fire and silvicultural strategies to offset climate effects and bolster resilience to fire, beetles, and drought.

Multi granularity based label propagation with active learning for semi-supervised classification

Semi-supervised learning (SSL) methods, which exploit both the labeled and unlabeled data, have attracted a lot of attention. One of the major categories of SSL methods, graph-based semi-supervised learning (GBSSL) learns labels of unlabeled data on an adjacency graph, where neighborhood sparse graph is often used to reduce computational complexity. However, the neighborhood size is difficult to set. Instead of assigning a concrete value of neighborhood size, we propose a new label propagation algorithm called multi granularity based label propagation (MGLP) and developed from the view of granular computing. In MGLP, labels of unlabeled data are learned by two classic label propagation processes with diverse neighborhood size k, where granular computing delivers a guiding strategy to leverage multiple level neighborhood information granules, and three-way decision acts as an active learning strategy to select the unlabeled data for further annotating. Through the iterative procedures of label propagating, data annotating and data subset updating, the ultimate pseudo label accuracy of unlabeled data may be higher. Theoretically, the accuracy of pseudo labels is enhanced in some scenarios. Experimentally, the results of simulation studies on ten benchmark datasets, show that the proposed method MGLP can rise pseudo labels accuracy by 8.6% than LP (label propagation), 6.5% than LNP (linear neighborhood propagation), 6.4% than LPSN (label propagation through sparse neighborhood), 4.5% than Adaptive-NP (adaptive neighborhood propagation) and 4.6% than CRLP (consensus rate-based label propagation). It also provides a novel way to annotate data.

Analog Circuit Soft Fault Diagnosis Based on Sparse Random Projections and K-Nearest Neighbor

Analog circuit fault diagnosis is a key problem in theory of circuit networks and has been investigated by many researchers in recent years. An approach based on sparse random projections (SRPs) and K-nearest neighbor (KNN) to the realization of analog circuit soft fault diagnosis has been presented in this paper. The proposed method uses the wavelet packet energy spectrum and sparse random projections to preprocess the time response for feature extraction. Then, the variables of the fault features are constructed, which are used to form the observation sequences of K-nearest neighbor classifier. K-nearest neighbor classifier is used to accomplish the fault diagnosis of analog circuit. In this paper, four-opamp biquad high-pass filter has been used as simulation example to verify the effectiveness of the proposed method. The simulations show that the proposed method offers higher correct fault location rate in analog circuit soft fault diagnosis application as compared with the other methods.

Phonological precision for word recognition in skilled readers.

According to the lexical quality hypothesis, differences in the orthographic, semantic, and phonological representations of words will affect individual reading performance. While several studies have focused on orthographic precision and semantic coherence, few have considered phonological precision. The present study used a suite of individual difference measures to assess which components of lexical quality contributed to competition resolution in a masked priming experiment. The experiment measured form priming for word and pseudoword targets with dense and sparse neighbourhoods in 84 university students. Individual difference measures of language and cognitive skills were also collected and a principal component analysis was used to group these data into components. The data showed that phonological precision and NHD interacted with form priming. In participants with high phonological precision, the direction of priming for word targets with sparse neighbourhoods was facilitatory, while the direction for those with dense neighbourhoods was inhibitory. In contrast, people with low phonological precision showed the opposite pattern, but the interaction was non-significant. These results suggest that the component of phonological precision is linked to lexical competition for word recognition and that access to the mental lexicon during reading is affected by differing levels of phonological processing.

Pupillometry reveals cognitive demands of lexical competition during spoken word recognition in young and older adults.

In most contemporary activation-competition frameworks for spoken word recognition, candidate words compete against phonological "neighbors" with similar acoustic properties (e.g., "cap" vs. "cat"). Thus, recognizing words with more competitors should come at a greater cognitive cost relative to recognizing words with fewer competitors, due to increased demands for selecting the correct item and inhibiting incorrect candidates. Importantly, these processes should operate even in the absence of differences in accuracy. In the present study, we tested this proposal by examining differences in processing costs associated with neighborhood density for highly intelligible items presented in quiet. A second goal was to examine whether the cognitive demands associated with increased neighborhood density were greater for older adults compared with young adults. Using pupillometry as an index of cognitive processing load, we compared the cognitive demands associated with spoken word recognition for words with many or fewer neighbors, presented in quiet, for young (n = 67) and older (n = 69) adult listeners. Growth curve analysis of the pupil data indicated that older adults showed a greater evoked pupil response for spoken words than did young adults, consistent with increased cognitive load during spoken word recognition. Words from dense neighborhoods were marginally more demanding to process than words from sparse neighborhoods. There was also an interaction between age and neighborhood density, indicating larger effects of density in young adult listeners. These results highlight the importance of assessing both cognitive demands and accuracy when investigating the mechanisms underlying spoken word recognition.

What Underlies the Deficit in Rapid Automatized Naming (RAN) in Adults with Dyslexia? Evidence from Eye Movements

ABSTRACT This eye-tracking study explored how phonological encoding and speech production planning for successive words are coordinated in adult readers with dyslexia (N = 22) and control readers (N = 25) during rapid automatized naming (RAN). Using an object-RAN task, we orthogonally manipulated the word-form frequency and phonological neighborhood density of the object names and assessed the effects on speech and eye movements and their temporal coordination. In both groups, there was a significant interaction between word frequency and neighborhood density: shorter fixations for dense than for sparse neighborhoods were observed for low- but not for high-frequency words. This finding does not suggest a specific difficulty in lexical phonological access in dyslexia. However, in readers with dyslexia only, these lexical effects percolated to the late processing stages, indicated by longer offset eye-speech lags. We close by discussing potential reasons for this finding, including suboptimal specification of phonological representations and deficits in attention control or in multi-item coordination.