Relevant Component Analysis Research Articles

Disentangled Explanations of Neural Network Predictions by Finding Relevant Subspaces.

Explainable AI aims to overcome the black-box nature of complex ML models like neural networks by generating explanations for their predictions. Explanations often take the form of a heatmap identifying input features (e.g. pixels) that are relevant to the model's decision. These explanations, however, entangle the potentially multiple factors that enter into the overall complex decision strategy. We propose to disentangle explanations by extracting at some intermediate layer of a neural network, subspaces that capture the multiple and distinct activation patterns (e.g. visual concepts) that are relevant to the prediction. To automatically extract these subspaces, we propose two new analyses, extending principles found in PCA or ICA to explanations. These novel analyses, which we call principal relevant component analysis (PRCA) and disentangled relevant subspace analysis (DRSA), maximize relevance instead of e.g. variance or kurtosis. This allows for a much stronger focus of the analysis on what the ML model actually uses for predicting, ignoring activations or concepts to which the model is invariant. Our approach is general enough to work alongside common attribution techniques such as Shapley Value, Integrated Gradients, or LRP. Our proposed methods show to be practically useful and compare favorably to the state of the art as demonstrated on benchmarks and three use cases.

Viable human brain microvessels for the study of aging and neurodegenerative diseases.

The brain microvasculature is altered in normal aging and in the presence of disease processes, such as neurodegeneration or ischemia, but there are few methods for studying living tissues. We now report that viable microvessels (MV) are readily isolated from brain tissue of subjects enrolled in studies of neurodegenerative diseases who undergo rapid autopsy (performed with <12h postmortem interval - PMI). We find that these MV retain their morphology and cellular components and are fairly uniform in size. Sufficient MV (~3-5000) are obtained from 3 to 4g of tissue to allow for studies of cellular composition as well as extracellular matrix (ECM). Using live/dead assays, these MV are viable for up to 5days in tissue culture media (2D) designed to support endothelial cells and up to 11days post-isolation in a 3-dimensional (3D) matrix (Low Growth Factor Matrigel™). Assays that measure the reducing potential of live cells \demonstrated that the majority of the MV maintain high levels of metabolic activity for a similar number of days as the live/dead assays. Functional cellular components (such as tight junctions and transporter proteins) and ECM of MV in tissue culture media, and to a lesser extent in 3D matrices, were readily visualized using immunofluorescence techniques. MV in tissue culture media are lysed and protein content analyzed, but MV in 3D matrix first require removal of the supporting matrix, which can confound the analysis of MV ECM. Finally, MV can be preserved in cryoprotective media, whereby over 50% retain their baseline viability upon thawing. In summary, we find that MV isolated from human brains undergoing rapid autopsy are viable in standard tissue culture for up to 5days and the timeframe for experiments can be extended up to 11days by use of a supportive 3D matrix. Viable human MV allow for temporal and spatial analysis of relevant cellular and ECM components that have implications for microvascular function in neurodegenerative diseases, vascular brain injury, and neurotrauma.

Cost of Pediatric Concurrent Hospice Care: An Economic Analysis of Relevant Cost Components, Review of the Literature, and Case Illustration.

Background: Concurrent care enables seriously ill pediatric Medicaid and Children's Health Insurance Program (CHIP) beneficiaries to continue curative treatments along with the supportive services usually associated with hospice care. Although a few studies have examined pediatric concurrent care, none has explored the economics of this care delivery approach for medically complex children. Objective: The purpose of this study was to identify the cost components relevant for an economic analysis of pediatric concurrent hospice care and demonstrate the use of the cost components in an economic case illustration. Methods: A scoping review of the hospice cost literature was conducted to identify article focused on hospice costs that were published in the journal literature from 2000 to 2019. A preliminary economic analysis was also conducted using a cost-minimization approach with data from a large mid-Atlantic pediatric hospice provider. We compared costs between concurrent versus standard hospice care for children. Results: Fifteen hospice cost articles were identified and no article on concurrent care costs. Most articles in the scoping review focused on hospice personnel costs, followed by supplies/equipment, and other. Personnel and costs were higher for standard hospice children compared with concurrent care, whereas concurrent care children had higher supplies/equipment costs. Conclusions: Identifying cost components are critical to economic analysis of pediatric concurrent care. These findings provide preliminary evidence about the difference in costs between concurrent and standard hospice care for children.

Quality relevant over-complete independent component analysis based monitoring for non-linear and non-Gaussian batch process

A large number of multivariate statistical methods have been applied to process monitoring, but conventional methods only extract limited feature information that often cannot effectively monitor the quality related changes of production characteristics in batch processes. In order to improve the effect of the process monitoring, this paper proposes a batch process monitoring method based on Multistage Over-complete Independent Component Analysis (OICA) algorithm. Firstly, the Affinity Propagation algorithm (AP) is used to divide the batch production process. Secondly, the extra quality information extracted by Partial Least Squares (PLS) algorithm is input into OICA algorithm. Finally, a monitoring model is established for process monitoring in each sub-stage. The effectiveness of the proposed method has been verified by comparing with the conventional methods in the fed-batch penicillin fermentation process.

Reconstruction of Simulated Magnetic Resonance Fingerprinting Using Accelerated Distance Metric Learning

Purpose: Magnetic Resonance Fingerprinting (MRF) is a novel framework that uses a random acquisition to acquire a unique tissue response, or fingerprint. Through a pattern-matching algorithm, every voxel-vise fingerprint is matched with a pre-calculated dictionary of simulated fingerprints to obtain MR parameters of interest. Currently, a correlation algorithm performs the MRF matching, which is time-consuming. Moreover, MRF suffers from highly undersampled k-space data, thereby reconstructed images have aliasing artifact, propagated to the estimated quantitative maps. We propose using a distance metric learning method as a matching algorithm and a Singular Value Decomposition (SVD) to compress the dictionary, intending to promote the accuracy of MRF and expedite the matching process. Materials and Methods: In this investigation, a distance metric learning method, called the Relevant Component Analysis (RCA) was used to match the fingerprints from the undersampled data with a compressed dictionary to create quantitative maps accurately and rapidly. An Inversion Recovery Fast Imaging with Steady-State (IR-FISP) MRF sequence was simulated based on an Extended Phase Graph (EPG) on a digital brain phantom. The performance of our work was compared with the original MRF paper. Results: Effectiveness of our method was evaluated with statistical analysis. Compared with the correlation algorithm and full-sized dictionary, this method acquires tissue parameter maps with more accuracy and better computational speed. Conclusion: Our numerical results show that learning a distance metric of the undersampled training data accompanied by a compressed dictionary improves the accuracy of the MRF matching and overcomes the computation complexity.

Quantification of oleacein and oleuropein aglycone in olive oil using deuterated surrogates by normal-phase ultra high performance liquid chromatography with quadrupole time-of-flight mass spectrometry.

An analytical method for the analysis of relevant secoiridoid-based components in olive oil, oleacein and oleuropein aglycone, is described using for the first time deuterated surrogates. 0.2g of sample was necessary to perform the analysis using liquid-liquid extraction and ultrasound-assisted extraction with a mixture of methanol/water (4:1, v/v). To avoid the formation of by-products, normal-phase ultra high performance liquid chromatography was chosen for the chromatographic separation. The selected mobile phase was a gradient mixture of tetrahydrofurane and hexane, and an ACE Excel 3 CN-ES column as stationary phase. The detection and quantification was performed with a SYNAPT G2-Si mass spectrometer. The calibration curves for oleacein and oleuropein aglycone were linear and quadratic, respectively. The validation was done at three levels of concentration. Relative errors from 0.1 to 10.5% and relative standard deviations lower than 9% were obtained. The method was applied to study different samples of olive oil.

Kernel Feature Extraction for Hyperspectral Image Classification Using Chunklet Constraints

A novel semi-supervised kernel feature extraction algorithm to combine an efficient metric learning method, i.e. relevant component analysis (RCA), and kernel trick is presented for hyperspectral imagery land-cover classification. This method obtains projection of the input data by learning an optimal nonlinear transformation via a chunklet constraints-based FDA criterion, and called chunklet-based kernel relevant component analysis (CKRCA). The proposed method is appealing as it constructs the kernel very intuitively for the RCA method and does not require any labeled information. The effectiveness of the proposed CKRCA is successfully illustrated in hyperspectral remote sensing image classification. Experimental results demonstrate that the proposed method can greatly improve the classification accuracy compared with traditional linear and conventional kernel-based methods.

Stevia rebaudiana Leaves: Effect of Drying Process Temperature on Bioactive Components, Antioxidant Capacity and Natural Sweeteners.

Stevia leaves are usually used in dried state and undergo the inevitable effect of drying process that changes the quality characteristics of the final product. The aim of this study was to assess temperature effect on Stevia leaves through analysis of relevant bioactive components, antioxidant capacity and content of natural sweeteners and minerals. The drying process was performed in a convective dryer at constant temperatures ranging from 30 to 80 °C. Vitamin C was determined in the leaves and as expected showed a decrease during drying proportional to temperature. Phenolics and flavonoids were also determined and were found to increase during drying below 50 °C. Antioxidant activity was determined by DPPH and ORAC assays, and the latter showed the highest value at 40 °C, with a better correlation with the phenolics and flavonoids content. The content of eight natural sweeteners found in Stevia leaves was also determined and an increase in the content of seven of the sweeteners, excluding steviol bioside, was found at drying temperature up to 50 °C. At temperatures between 60 and 80 °C the increase in sweeteners content was not significant. Stevia leaves proved to be an excellent source of antioxidants and natural sweeteners.

A Thermo-Hydraulic Tool for Automatic Virtual Hazop Evaluation

Abstract Development of complex lubrication systems in the Oil&amp;Gas industry has reached high levels of competitiveness in terms of requested performances and reliability. In particular, the use of HazOp (acronym of Hazard and Operability) analysis represents a decisive factor to evaluate safety and reliability of plants. The HazOp analysis is a structured and systematic examination of a planned or existing operation in order to identify and evaluate problems that may represent risks to personnel or equipment. In particular, P&amp;ID schemes (acronym of Piping and Instrument Diagram according to regulation in force ISO 14617) are used to evaluate the design of the plant in order to increase its safety and reliability in different operating conditions. The use of a simulation tool can drastically increase speed, efficiency and reliability of the design process. In this work, a tool, called TTH lib (acronym of Transient Thermal Hydraulic Library) for the 1-D simulation of thermal hydraulic plants is presented. The proposed tool is applied to the analysis of safety relevant components of compressor and pumping units, such as lubrication circuits. Opposed to the known commercial products, TTH lib has been customized in order to ease simulation of complex interactions with digital logic components and plant controllers including their sensors and measurement systems. In particular, the proposed tool is optimized for fixed step execution and fast prototyping of Real Time code both for testing and production purposes. TTH lib can be used as a standard SimScape-Simulink library of components optimized and specifically designed in accordance with the P&amp;ID definitions. Finally, an automatic code generation procedure has been developed, so TTH simulation models can be directly assembled from the P&amp;ID schemes and technical documentation including detailed informations of sensor and measurement system.

Physio-visual data fusion for emotion recognition

Several approaches have been proposed to recognize human emotions based on facial expressions or physiological signals, relatively rare work has been done to fuse these two, and other, modalities to improve the accuracy and robustness of the emotion recognition system. In this paper, we propose two methods based on feature-level and decision-level to fuse facial and physiological modalities. At feature-level fusion, we have tested the mutual information approach for selecting the most relevant and principal component analysis to reduce their dimensionality. For decision-level fusion, we have implemented two methods; the first is based on voting process and the second is based on dynamic Bayesian Networks. The system is validated using data obtained through an emotion elicitation experiment based on the International Affective Picture System. Results show that feature-level fusion is better than decision-level fusion.

A Feature-Metric-Based Affinity Propagation Technique for Feature Selection in Hyperspectral Image Classification

Relevant component analysis has shown effective in metric learning. It finds a transformation matrix of the feature space using equivalence constraints. This paper explores this idea for constructing a feature metric (FM) and develops a novel semisupervised feature-selection technique for hyperspectral image classification. Two feature measures referred to as band correlation metric (BCM) and band separability metric (BSM) are derived for the FM. The BCM can measure the spectral correlation among the bands, while the BSM can assess the class discrimination capability of a single band. The proposed feature-metric-based affinity propagation (AP) (FM-AP) technique utilizes exemplar-based clustering, i.e., AP, to group bands from original spectral channels with the FM. Experimental results are conducted on two hyperspectral images and show the advantages of the proposed technique over traditional feature-selection methods.

A New One-class Classifier: Relevant Component Analysis Data Description

Abstract It is known that Principal Component Analysis Data Description (PCADD) as a one-class classifier estimates the mean and covariance matrix of the target class objects and uses the Mahalanobis distance so as to fit the target class. However, PCADD just considers the global information of the target class and ignores the local information. In this paper, we introduce the local information into the one-class classifier design and develop a new one-class classifier called Relevant Component Analysis Data Description (RCADD). Concretely, we first divide the target class into some chunklets and each chunklet is made up of the similar objects. Then, the presented RCADD applies the RCA transformation onto the original target space. Finally, the RCADD computes the difference between the original object and the transformation of that object onto the subspace so as to check if the object belongs to the target class. The experimental results demonstrate that the proposed RCADD has a superior performance to the PCADD in terms of classification.

Generalized sparse metric learning with relative comparisons

The objective of sparse metric learning is to learn a distance measure from a set of data in addition to finding a low-dimensional representation. Despite demonstrated success, the performance of existing sparse metric learning approaches is usually limited because the methods assumes certain problem relaxations or they target the SML objective indirectly. In this paper, we propose a Generalized Sparse Metric Learning method. This novel framework offers a unified view for understanding many existing sparse metric learning algorithms including the Sparse Metric Learning framework proposed in (Rosales and Fung ACM International conference on knowledge discovery and data mining (KDD), pp 367–373, 2006), the Large Margin Nearest Neighbor (Weinberger et al. in Advances in neural information processing systems (NIPS), 2006; Weinberger and Saul in Proceedings of the twenty-fifth international conference on machine learning (ICML-2008), 2008), and the D-ranking Vector Machine (D-ranking VM) (Ouyang and Gray in Proceedings of the twenty-fifth international conference on machine learning (ICML-2008), 2008). Moreover, GSML also establishes a close relationship with the Pairwise Support Vector Machine (Vert et al. in BMC Bioinform, 8, 2007). Furthermore, the proposed framework is capable of extending many current non-sparse metric learning models to their sparse versions including Relevant Component Analysis (Bar-Hillel et al. in J Mach Learn Res, 6:937–965, 2005) and a state-of-the-art method proposed in (Xing et al. Advances in neural information processing systems (NIPS), 2002). We present the detailed framework, provide theoretical justifications, build various connections with other models, and propose an iterative optimization method, making the framework both theoretically important and practically scalable for medium or large datasets. Experimental results show that this generalized framework outperforms six state-of-the-art methods with higher accuracy and significantly smaller dimensionality for seven publicly available datasets.

Relevant principal component analysis applied to the characterisation of Portuguese heather honey

The main purpose of this study was the characterisation of ‘Serra da Lousã’ heather honey by using novel statistical methodology, relevant principal component analysis, in order to assess the correlations between production year, locality and composition. Herein, we also report its chemical composition in terms of sugars, glycerol and ethanol, and physicochemical parameters. Sugars profiles from ‘Serra da Lousã’ heather and ‘Terra Quente de Trás-os-Montes’ lavender honeys were compared and allowed the discrimination: ‘Serra da Lousã’ honeys do not contain sucrose, generally exhibit lower contents of turanose, trehalose and maltose and higher contents of fructose and glucose. Different localities from ‘Serra da Lousã’ provided groups of samples with high and low glycerol contents. Glycerol and ethanol contents were revealed to be independent of the sugars profiles. These data and statistical models can be very useful in the comparison and detection of adulterations during the quality control analysis of ‘Serra da Lousã’ honey.

Extending the relevant component analysis algorithm for metric learning using both positive and negative equivalence constraints

Relevant component analysis (RCA) is a recently proposed metric learning method for semi-supervised learning applications. It is a simple and efficient method that has been applied successfully to give impressive results. However, RCA can make use of supervisory information in the form of positive equivalence constraints only. In this paper, we propose an extension to RCA that allows both positive and negative equivalence constraints to be incorporated. Experimental results show that the extended RCA algorithm is effective.

A COMPARISON OF K-MEANS AND FUZZY C-MEANS USING BACKGROUND KNOWLEDGE

Relevant Component Analysis has been introduced recently as a way to incorporate a priori information, such as class or preference information, that may exist for a given data set. The method uses this information to define a new Mahalanobis distance metric on the data space. The purpose of the present paper is to investigate the difference, if any, that Relevant Component Analysis makes when applied in conjunction with the clustering algorithms of k-means or fuzzy c-means. Results are given for six standard data sets.

Boosting the distance estimation: Application to the K-Nearest Neighbor Classifier

In this work we introduce a new distance estimation technique by boosting and we apply it to the K-Nearest Neighbor Classifier ( K-NN). Instead of applying AdaBoost to a typical classification problem, we use it for learning a distance function and the resulting distance is used into K-NN. The proposed method (Boosted Distance with Nearest Neighbor) outperforms the AdaBoost classifier when the training set is small. It also outperforms the K-NN classifier used with several different distances and the distances obtained with other estimation methods such as Relevant Component Analysis (RCA) [Duda, R.O., Hart, P.E., Stock, D.G., 2001. Pattern Classification, John Wiley and Sons Inc.]. Furthermore, our distance estimation performs dimension-reduction, being much more efficient in terms of classification accuracy than classical techniques such as PCA, LDA, and NDA. The method has been thoroughly tested on 13 standard databases from the UCI repository, a standard gender recognition database and the MNIST database.

Temporal Analysis of Relevant and Irrelevant Behavior Components in Partial Reinforcement and Extinction

Several time measures describing the essential or relevant and the irrelevant behavior components in a water-reinforced lever-pressing situation were studied in 30 rats as a function of four reinforcement ratios (1:1, 4:1, 8:1 and 16:1) and of extinction procedure. Although the traditional frequency measures were affected in the expected way, the distributions of time taken to perform the relevant components of the lever-pressing response were minimally affected by the transitions either from one ratio to another or from any ratio to the extinction procedure. This was contrasted by marked changes in the incidence of irrelevant activities (sniffing and grooming). The implications of these findings for the current views of the nature of partial-reinforcement and extinction effects are discussed.