Locally Linear Embedding Research Articles

Unsupervised EEG preictal interval identification in patients with drug-resistant epilepsy

Typical seizure prediction models aim at discriminating interictal brain activity from pre-seizure electrographic patterns. Given the lack of a preictal clinical definition, a fixed interval is widely used to develop these models. Recent studies reporting preictal interval selection among a range of fixed intervals show inter- and intra-patient preictal interval variability, possibly reflecting the heterogeneity of the seizure generation process. Obtaining accurate labels of the preictal interval can be used to train supervised prediction models and, hence, avoid setting a fixed preictal interval for all seizures within the same patient. Unsupervised learning methods hold great promise for exploring preictal alterations on a seizure-specific scale. Multivariate and univariate linear and nonlinear features were extracted from scalp electroencephalography (EEG) signals collected from 41 patients with drug-resistant epilepsy undergoing presurgical monitoring. Nonlinear dimensionality reduction was performed for each group of features and each of the 226 seizures. We applied different clustering methods in searching for preictal clusters located until 2 h before the seizure onset. We identified preictal patterns in 90% of patients and 51% of the visually inspected seizures. The preictal clusters manifested a seizure-specific profile with varying duration (22.9 ± 21.0 min) and starting time before seizure onset (47.6 ± 27.3 min). Searching for preictal patterns on the EEG trace using unsupervised methods showed that it is possible to identify seizure-specific preictal signatures for some patients and some seizures within the same patient.

Surface Approximation by Means of Gaussian Process Latent Variable Models and Line Element Geometry

The close relation between spatial kinematics and line geometry has been proven to be fruitful in surface detection and reconstruction. However, methods based on this approach are limited to simple geometric shapes that can be formulated as a linear subspace of line or line element space. The core of this approach is a principal component formulation to find a best-fit approximant to a possibly noisy or impartial surface given as an unordered set of points or point cloud. We expand on this by introducing the Gaussian process latent variable model, a probabilistic non-linear non-parametric dimensionality reduction approach following the Bayesian paradigm. This allows us to find structure in a lower dimensional latent space for the surfaces of interest. We show how this can be applied in surface approximation and unsupervised segmentation to the surfaces mentioned above and demonstrate its benefits on surfaces that deviate from these. Experiments are conducted on synthetic and real-world objects.

Model order reduction techniques to identify submarining risk in a simplified human body model

This work investigates linear and non-linear parametric reduced order models (ROM) capable of replacing computationally expensive high-fidelity simulations of human body models (HBM) through a non-intrusive approach. Conventional crash simulation methods pose a computational barrier that restricts profound analyses such as uncertainty quantification, sensitivity analysis, or optimization studies. The non-intrusive framework couples dimensionality reduction techniques with machine learning-based surrogate models that yield a fast responding data-driven black-box model. A comparative study is made between linear and non-linear dimensionality reduction techniques. Both techniques report speed-ups of a few orders of magnitude with an accurate generalization of the design space. These accelerations make ROMs a valuable tool for engineers.

Linear dimensionality reduction method based on topological properties

Dimensionality reduction is an important data preprocessing technique that has been extensively studied in machine learning and data mining. Locality Preserving Projection (LPP) is a widely used linear unsupervised dimensionality reduction method, which maps high-dimensional data into low-dimensional subspace through linear transformation. Although various variants of LPP have been proposed to tackle different drawbacks of LPP, it is identified in this article that LPP does not possess the important topological property of translation invariance, that is, the linear transformation given by LPP is strongly related to the relative position between the data and the origin of the coordinate system. In this article, we theoretically analyze the reason why this drawback exists in LPP and propose to resolve it by introducing a kind of centralization to the model. Moreover, as topological properties are prominent information to characterize the structure of the data, this article proposes a further improvement of LPP to maintain topological connectivity of data after dimensionality reduction. Experiments on multiple synthetic and real-world datasets show that the new model incorporating topological properties outperforms not only the original LPP model but also several other classic linear or non-linear dimensionality reduction methods.

Protein Interaction Prediction Method Based on Feature Engineering and XGBoost

Human protein interaction prediction studies occupy an important place in systems biology. The understanding of human protein interaction networks and interactome will provide important insights into the regulation of developmental, physiological and pathological processes. In this study, we propose a method based on feature engineering and integrated learning algorithms to construct protein interaction prediction models. Principal Component Analysis (PCA) and Locally Linear Embedding (LLE) dimensionality reduction methods were used to extract sequence features from the 174-dimensional human protein sequence vector after Normalized Difference Sequence Feature (NDSF) encoding, respectively. The classification performance of three integrated learning methods (AdaBoost, Extratrees, XGBoost) applied to PCA and LLE features was compared, and the best combination of parameters was found using cross-validation and grid search methods. The results show that the classification accuracy is significantly higher when using the linear dimensionality reduction method PCA than the nonlinear dimensionality reduction method LLE. the classification with XGBoost achieves a model accuracy of 99.2%, which is the best performance among all models. This study suggests that NDSF combined with PCA and XGBoost may be an effective strategy for classifying different human protein interactions.

A Survey of Manifold Learning and Its Applications for Multimedia

Manifold learning is an emerging research domain of machine learning and helps to overcome the limits of standard neural networks which are restricted to Euclidean spaces. In this work, we give an introduction into manifold learning and how it is employed for important application fields in multimedia like similarity search, image classification, synthesis & enhancement, video analysis, 3D data processing and nonlinear dimension reduction. Furthermore, we present available open source software frameworks for manifold learning. One can see from this survey that manifold learning has a lot of potential for computer vision and multimedia.

DBSCAN-Based Automatic De-Duplication for Software Quality Inspection Data

Software quality inspection will generate too much data, and removing duplicate data can improve the efficiency of software quality inspection. This paper studies the automatic de-duplication method of software quality inspection data based on density-based spatial clustering of applications with noise (DBSCAN) clustering. Intelligent optimization algorithm is used to generate software quality inspection data by initializing individuals, calculating fitness function value, improving individuals and splitting individuals that meet the conditions. Local linear embedding algorithm is selected to extract software quality inspection data features by searching neighborhood points, calculating reconstruction weight and projection vector. The extracted features are used to select DBSCAN multi-density clustering algorithm of regional division, and the automatic de-duplication of software quality inspection data is realized by grid division, data bin dividing and grid merging. The experimental results show that the precision and recall of this method are higher than 99%, and the resource consumption rate is low, which can effectively improve the efficiency of software quality inspection.

Insightful Visualization of Remote Sensing Images

Numerous dimensions exist in remote sensing imagery (RSI) dataset, necessitating visualization to display and enable human analysis. In order to visualize, dimension reduction (DR) is performed to decrease the dimensionality to two or three. This letter proposes a novel DR technique called nonlinear local linear embedding (NLLE), where the objective is to obtain an accurate visualization of RSI datasets. NLLE reduces the high dimension of large datasets to a space with low dimensionality, preserving the neighborhood relationship between points in the original space. When compared to other well-known DR techniques, NLLE’s flexible procedure increases the likelihood of achieving effective RSI visualization.

SVMWPE-HLLE Based Fault Diagnosis Approach for Rolling Bearing

Multi-scale permutation entropy (MPE) is an analytical method describing the complexity of time series, which has been applied to the fault diagnosis of rolling bearings. To solve the problems of MPE in coarse-grained process and permutation entropy calculation, multi-scale weighted permutation entropy based on sliding variance (SVMWPE) was proposed in this paper. By analyzing WGN signal and 1/f noise signal, the parameter selections for SVMWPE were studied, and the stability and superiority were investigated by comparing SVMWPE with MPE, MWPE, and SVMPE. The high-dimensional matrix obtained by MPE feature extraction to pattern recognition was solved by introducing Hessian local linear embedding (HLLE) dimension reduction method, and the feature extraction method based on SVMWPE-HLLE was proposed. The clustering effect was studied by comparing SVMWPE-HLLE with SVMWPE-LLE through the analysis of three simulation signals. Fault diagnosis method for rolling bearing was proposed by combining SVMWPE-HLLE with extreme learning machine (ELM), which was applied to two experimental cases of rolling bearings for analysis. The experimental results showed that the proposed method can realize intelligent diagnosis of different fault types and degrees of rolling bearing, and the fault recognition rate of the proposed method was higher than other methods.

A Guide on Analyzing Flow Cytometry Data Using Clustering Methods and Nonlinear Dimensionality Reduction (tSNE or UMAP).

Flow cytometry has been used for the last two decades to identify which immune cell subsets diapedese from the periphery into the brain parenchyma following injuries, including ischemic and hemorrhagic stroke. Recent developments have moved the analysis of high-parameter flow cytometry data sets from the traditional analysis method of manual gating to using unbiased analyses to improve scientific rigor. This chapter gives a step-by-step guide on using modern computational approaches to analyze complex flow cytometry data sets in FlowJo™ Software v10. The section will describe pre-processing and outline the steps needed to perform unsupervised clustering of your data set in addition to using nonlinear dimensionality reduction for visualizing your analysis. While these methods can identify long-term neuroinflammatory responses after stroke, the methods could be applied to a variety of flow cytometry data sets.

Neural networks with dimensionality reduction for predicting temperature change due to plastic deformation in a cold rolling simulation

Abstract Cold rolling involves large deformation of the workpiece leading to temperature increase due to plastic deformation. This process is highly nonlinear and leads to large computation times to fully model the process. This paper describes the use of dimension-reduced neural networks (DR-NNs) for predicting temperature changes due to plastic deformation in a two-stage cold rolling process. The main objective of these models is to reduce computational demand, error, and uncertainty in predictions. Material properties, feed velocity, sheet dimensions, and friction models are introduced as inputs for the dimensionality reduction. Different linear and nonlinear dimensionality reduction methods reduce the input space to a smaller set of principal components. The principal components are fed as inputs to the neural networks for predicting the output temperature change. The DR-NNs are compared against a standalone neural network and show improvements in terms of lower computational time and prediction uncertainty.

Emotion recognition algorithm of basketball players based on deep learning

Aiming at the problems of traditional methods of emotion recognition accuracy, long recognition time and low recognition rate, a basketball player emotion recognition algorithm based on deep learning is proposed. Based on the Emotic dataset, a basketball remote mobilisation emotion recognition dataset is constructed to realise emotion classification. The LBP method is used to extract the facial expression features in the dataset, and the KDIsomap algorithm is used to perform nonlinear dimensionality reduction on the features according to the feature extraction results. According to the deep learning algorithm, the SVM classifier is combined with the KNN classification to form an SVM-KNN classifier to recognise the emotions of basketball players. Experimental results show that the shortest recognition time of the proposed algorithm is only 4.38 s, the highest recognition accuracy rate reaches 94.2%, and the recognition rate is high, indicating that the algorithm has a certain effectiveness.

Mapping of structural arrangement of cells and collective calcium transients: an integrated framework combining live cell imaging using confocal microscopy and UMAP-assisted HDBSCAN-based approach.

Live cell calcium (Ca2+) imaging is one of the important tools to record cellular activity during in vitro and in vivo preclinical studies. Specially, high-resolution microscopy can provide valuable dynamic information at the single cell level. One of the major challenges in the implementation of such imaging schemes is to extract quantitative information in the presence of significant heterogeneity in Ca2+ responses attained due to variation in structural arrangement and drug distribution. To fill this gap, we propose time-lapse imaging using spinning disk confocal microscopy and machine learning-enabled framework for automated grouping of Ca2+ spiking patterns. Time series analysis is performed to correlate the drug induced cellular responses to self-assembly pattern present in multicellular systems. The framework is designed to reduce the large-scale dynamic responses using uniform manifold approximation and projection (UMAP). In particular, we propose the suitability of hierarchical DBSCAN (HDBSCAN) in view of reduced number of hyperparameters. We find UMAP-assisted HDBSCAN outperforms existing approaches in terms of clustering accuracy in segregation of Ca2+ spiking patterns. One of the novelties includes the application of non-linear dimension reduction in segregation of the Ca2+ transients with statistical similarity. The proposed pipeline for automation was also proved to be a reproducible and fast method with minimal user input. The algorithm was used to quantify the effect of cellular arrangement and stimulus level on collective Ca2+ responses induced by GPCR targeting drug. The analysis revealed a significant increase in subpopulation containing sustained oscillation corresponding to higher packing density. In contrast to traditional measurement of rise time and decay ratio from Ca2+ transients, the proposed pipeline was used to classify the complex patterns with longer duration and cluster-wise model fitting. The two-step process has a potential implication in deciphering biophysical mechanisms underlying the Ca2+ oscillations in context of structural arrangement between cells.

Measure of Similarity between GMMs Based on Geometry-Aware Dimensionality Reduction

Gaussian Mixture Models (GMMs) are used in many traditional expert systems and modern artificial intelligence tasks such as automatic speech recognition, image recognition and retrieval, pattern recognition, speaker recognition and verification, financial forecasting applications and others, as simple statistical representations of underlying data. Those representations typically require many high-dimensional GMM components that consume large computing resources and increase computation time. On the other hand, real-time applications require computationally efficient algorithms and for that reason, various GMM similarity measures and dimensionality reduction techniques have been examined to reduce the computational complexity. In this paper, a novel GMM similarity measure is proposed. The measure is based on a recently presented nonlinear geometry-aware dimensionality reduction algorithm for the manifold of Symmetric Positive Definite (SPD) matrices. The algorithm is applied over SPD representations of the original data. The local neighborhood information from the original high-dimensional parameter space is preserved by preserving distance to the local mean. Instead of dealing with high-dimensional parameter space, the method operates on much lower-dimensional space of transformed parameters. Resolving the distance between such representations is reduced to calculating the distance among lower-dimensional matrices. The method was tested within a texture recognition task where superior state-of-the-art performance in terms of the trade-off between recognition accuracy and computational complexity has been achieved in comparison with all baseline GMM similarity measures.

EEG signals classification using a new radial basis function neural network and jellyfish meta-heuristic algorithm.

The online version contains supplementary material available at 10.1007/s12065-022-00802-2.

ETHNICITY-SPECIFIC EXTREME LONGEVITY VARIANTS IN A CONSORTIUM OF FOUR CENTENARIAN STUDIES

Abstract It has been shown that some longevity variants including the APOE variants have ethnicity-specific effects on EL within European ethnicities. The goal of the present study is to identify genetic variants whose effects vary by ethnicity by conducting a genome-wide association study of extreme longevity (EL: defined as living past the age at which less than 1% individuals from the 1900 - 1920 birth year cohorts survived) that includes the SNP-by-ethnicity interaction terms using a consortium of four centenarian studies: the New England Centenarian Study, the Long Life Family Study, the Southern Italian Centenarian Study, and the Longevity Gene Project. We used the Uniform Manifold Approximation and Projection (UMAP), a non-linear dimension reduction technique, to identify distinct ethnic clusters. The UMAP analysis revealed four distinct ethnic groups (Danish, Italian, Ashkenazi Jewish, and middle European) in the aggregated data set with 2223 cases and 5673 controls. Using a mixed effects logistic model with SNP-by-ethnicity interaction terms, we found 29 loci, in which the test of any interaction effect produced p< 10-5. The results showed that some variants had ethnicity-specific effects on EL. We sought for replication in two independent studies of longevity: the Danish Longevity Study and the Italian Longevity Study. In the Italian Longevity Study, rs7907949 (PFKP), rs11667516 (intergenic: RAB11B;MARCHF2), rs13245505 (intergenic: ZC3HC1;KLHDC10) replicated with a nominal significance level of 0.05. In the Danish Longevity Study, rs79853795 (PLCB1) and rs4072601 (SLC39A11) replicated. Future drug development should account for ethnic-specific differences in the genetic effects for higher efficacy for more diverse populations.

Hierarchical unimodal processing within the primary somatosensory cortex during a bimodal detection task

Do sensory cortices process more than one sensory modality?To answer these questions, scientists have generated a wide variety of studies at distinct space-time scales in different animal models, and often shown contradictory conclusions. Some conclude that this process occurs in early sensory cortices, but others that this occurs in areas central to sensory cortices. Here, we sought to determine whether sensory neurons process and encode physical stimulus properties of different modalities (tactile and acoustic). Forthis, we designed a bimodal detection task where the senses of touch and hearing compete from trial to trial. Two Rhesus monkeys performed this novel task, while neural activity was recorded in areas 3b and 1 of the primary somatosensory cortex (S1). Weanalyzed neurons' coding properties and variability, organizing them by their receptive field's position relative to the stimulation zone. Our results indicate that neurons of areas 3b and 1 are unimodal, encoding only the tactile modality in both the firing rate and variability. Moreover, we found that neurons in area 3b carried more information about the periodic stimulus structure than those in area 1, possessed lower response and coding latencies, and had a lower intrinsic time scale. In sum, these differences reveal a hidden processing-based hierarchy. Finally, using a powerful nonlinear dimensionality reduction algorithm, we show that the activity from areas 3b and 1 can be separated, establishing a clear division in the functionality of these two subareas of S1.

Geometric learning for computational mechanics Part II: Graph embedding for interpretable multiscale plasticity

The history-dependent behaviors of classical plasticity models are often driven by internal variables evolved according to phenomenological laws. The difficulty to interpret how these internal variables represent a history of deformation, the lack of direct measurement of these internal variables for calibration and validation, and the weak physical underpinning of those phenomenological laws have long been criticized as barriers to creating realistic models. In this work, geometric machine learning on graph data (e.g. finite element solutions) is used as a means to establish a connection between nonlinear dimensional reduction techniques and plasticity models. Geometric learning-based encoding on graphs allows the embedding of rich time-history data onto a low-dimensional Euclidean space such that the evolution of plastic deformation can be predicted in the embedded feature space. A corresponding decoder can then convert these low-dimensional internal variables back into a weighted graph such that the dominating topological features of plastic deformation can be observed and analyzed.

Local Linear Embedding with Adaptive Neighbors

Dimensionality reduction is one of the most important techniques in the field of data mining. It embeds high-dimensional data into a low-dimensional vector space while keeping the main information as much as possible. Locally Linear Embedding (LLE) as a typical manifold learning algorithm computes neighborhood preserving embeddings of high-dimensional inputs. Based on the thought of LLE, we propose a novel unsupervised dimensionality reduction model called Local Linear Embedding with Adaptive Neighbors (LLEAN). To achieve a desirable dimensionality reduction result, we impose adaptive neighbor strategy and adopt a projection matrix to project data into an optimal subspace. The relationship between every pair-wise data is investigated to help reveal the data structure. Augmented Lagrangian Multiplier (ALM) is devised in optimization procedure to effectively solve the proposed objective function. Comprehensive experiments on toy data and benchmark datasets have been done and the results show that LLEAN outperforms other state-of-the-art dimensionality reduction methods.

Dual-weight local linear embedding algorithm based on adaptive neighborhood

Local linear embedding (LLE) algorithm mainly depends on local structures to extract significant features; however, the local structures are sensitive to the selection of neighborhood parameters. To solve this problem, the dual-weight local linear embedding algorithm based on adaptive neighborhood (AN-DWLLE) is proposed. First, the neighborhood parameter is adaptively estimated by computing the distances between the inquiry point and the centers of the inquiry point’s candidate neighborhoods. Then, the neighbor sequence and the local linear structures are integrated to construct the robust local structure that can be used to extract the significant features. A large number of experiments are performed on two bearing data sets. The experimental results illustrate that, compared with other related methods, the AN-DWLLE algorithm can extract more salient features and has greater dimensionality reduction effect.