Search

Tuning the values of kernel parameters plays a vital role in the performance of kernel methods. Kernel path algorithms have been proposed for several important learning algorithms, including support vector machine and kernelized Lasso, which can fit the piecewise nonlinear solutions of kernel methods with respect to the kernel parameter in a continuous space. Although the error path algorithms have been proposed to ensure that the model with the minimum cross validation (CV) error can be found, which is usually the ultimate goal of model selection, they are limited to piecewise linear solution paths. To address this problem, in this article, we extend the classic error path algorithm to the nonlinear kernel solution paths and propose a new kernel error path algorithm (KEP) that can find the global optimal kernel parameter with the minimum CV error. Specifically, we first prove that error functions of binary classification and regression problems are piecewise constant or smooth w.r.t. the kernel parameter. Then, we propose KEP for support vector machine and kernelized Lasso and prove that it guarantees to find the model with the minimum CV error within the whole range of kernel parameter values. Experimental results on various datasets show that our KEP can find the model with minimum CV error with less time consumption. Finally, it would have better generalization error on the test set, compared with grid search and random search.

Introduction: Autonomic dysfunction is common in chronic obstructive pulmonary disease (COPD), and worse autonomic function may be a marker of risk for acute exacerbations of COPD (AECOPD). Heart rate variability (HRV) is a measure of autonomic function. Our objective was to test whether lower (worse) HRV is a risk factor for AECOPD. Methods: We measured standard deviation of normal RR intervals (SDNN) and root mean square of successive RR interval differences (RMSSD) on 10-second electrocardiograms (ECGs) performed at screening and day 42 in participants in the Beta Blockers for the Prevention of Acute Exacerbations of COPD trial ( BLOCK-COPD), a placebo-controlled trial of metoprolol for prevention of AECOPD. We used Cox-proportional hazards models to test if these HRV measures were associated with risk of any AECOPD, and separately, hospitalized AECOPD. We tested associations using baseline HRV measures and incorporating HRV measures from day 42 as a time-varying covariate. We also tested for interactions with metoprolol assignment. Results: Of 532 trial participants, 529 (forced expiratory volume in 1 second [FEV(1) ]41 ± 16.3 % predicted) were included in this analysis. We did not find a significant association between HRV measures and risk of AECOPD when all participants were analyzed together. There was a significant interaction between RMSSD and assignment to metoprolol on time to first hospitalized AECOPD; in the placebo group greater RMSSD was associated with a lower risk of hospitalized AECOPD (adjusted hazard ratio0.71, 95% confidence interval: 0.52 to 0.96, per 10 ms increase) but there was no association in the metoprolol group. Conclusions: Autonomic dysfunction as measured by HRV may be a risk factor for AECOPD. Future studies should analyze longer HRV recordings and their performance in broader samples of people with COPD, including those on beta-blockers.

High-density aluminum foam can provide higher stiffness and absorb more energy during the impact Obtaining the constitutive law of such foam requires tri-axial tests with very high pressure, where difficulty may arise because the hydrostatic pressure can reach more than 30 MPa. In this paper, instead of using tri-axial tests, we proposed three easier tests—tension, compression and shear to obtain the parameters of constitutive model (the Deshpande–Fleck model). To verify the constitutive model both statically and dynamically, we carried out additional triaxial tests and direct impact tests, respectively. Based on the derived model, we performed finite element simulation to study the impact response of the present foam. By dimensional analysis, we proposed an empirical equation for a non-dimensional impact time $${\bar{t}}_{\mathrm {d}}$$ , the impact time divided by the time required for plastic wave travelling from the impact surface to the bottom surface, to determine the deformation characteristic of the aluminum foam after impact. For the case with $${\bar{t}}_{\mathrm {d}}\le 1$$ , the deformation tends to exhibit a shock-type characteristic, while for the case with $${\bar{t}}_{\mathrm {d}}>5$$ , the deformation tends to exhibit an upsetting-type characteristic.

The cell plasma membrane suffers structural disruptions from both daily environmental stresses and trauma. Rapid loss of cell viability occurs if membrane integrity is not rapidly restored. Physiological membrane sealing involves alteration of local intermolecular thermodynamics that is manifested by changes in membrane tension which precede reassembly of the membrane planar bilayer structure. Certain block copolymer surfactants, including poloxamer 188 (P188), have been proven to seal-disrupted cell membranes. However, the specific molecular mechanics of poloxamer-mediated membrane sealing remains a target of investigation. A decrease in membrane tension precedes membrane sealing by natural intrinsic cell sealing processes. The effect of P188 on the quasistatic membrane tension of Madin-Darby canine kidney (MDCK) and Swiss 3T3 fibroblasts cells under normal and saponin-permeabilized conditions was measured using laser optical tweezer (LOT)-extracted membrane tethers. The tether trap length of saponin-permeabilized MDCK cell membranes decreased from an uninjured control of 11.28 ± 1.1 μm to 6.43 ± 0.67 μm. Treatment with P188 (0.2 mM) significantly increased the tether trap length to 9.69 ± 1.0 μm (p < 0.05) while the control polymer, polyethylene glycol (0.2 mM) resulted in tether trap length of 7.02 ± 0.73 μm that was not significantly different. Similar observations were made in the saponin-permeabilized fibroblasts. Corresponding fluorescence cell viability assays revealed that P188-treated cells had a higher survival rate. Thus, surfactant copolymer membrane sealing restores the membrane integrity by decreasing the membrane tension. Cells quickly become non-viable when plasma membrane integrity is lost. Restoring or healing the cell membrane requires alteration in the forces that stabilize the membrane structure. The membrane defect healing processes are preceded by a decrease in membrane interfacial tension. This study demonstrated that amphiphilic block copolymer surfactants, such as poloxamer 188 (P188), reduce membrane tension and promote cell survival. Using laser optical tweezers, this work quantified the effect of surfactant copolymer-catalyzed sealing. Our work utilized laser optical tweezers to demonstrate that copolymer-catalyzed membrane sealing corresponds with a decrease in membrane tension, much like the physiologic response to membrane injury.

Trauma-mediated peripheral nerve axon injury invariably involves structural alteration of the axonal plasmalemma cell membrane. Increased axonal membrane permeability to electrolytes results in metabolic stress and axonal degeneration if the axon does not seal. Poloxamer 188 (P188) is a linear multiblock copolymer composed of a hydrophobic polyoxypropylene (POP) block in the middle with polyethylene glycol (PEG) blocks on both ends. The central POP block concentrates at sites of cell membrane disruption and catalyzes sealing of disrupted cellular membranes. We tested the hypothesis that intravenously administered P188 (9800 Da, 0.2 mM) can accelerate distal axon recovery after sciatic nerve crush injury using an established axonotemetic model. A crush injury was performed on the sciatic nerve of thirty female Sprague-Dawley rats. Rats were randomized to receive either 0.2 mM P188 or Dextran, administered 1 h after crush injury. Compound nerve action potentials (CNAP) and density of neurofilament distal to the point of injury were analyzed on post-crush day (PCD) 4, 14, and 21. A significant improvement in axonal conduction for animals treated with P188 was observed for PCD 4 and PCD 14 (p

Electrical stun devices (ESDs) serve a basic role in law enforcement and provide an alternative to lethal options for target control by causing electromuscular incapacitation (EMI). A fundamental concern is the adverse health consequences associated with their use. The capability of EMI electric field pulses to disrupt skeletal muscle cells (i.e. rhabdomyolysis) was investigated over the operational range commonly used in commercial EMI devices. Functional and structural alteration and recovery of muscle and nerve tissue were assessed. In an anesthetized swine model, the left thigh was exposed to 2 min of electrical pulses, using a commercially available ESD or a custom-made EMI signal power amplifier. Serum creatinine phosphokinase (CPK), troponin, aspartate aminotransferase (AST), and lactate dehydrogenase (LDH) levels were monitored intermittently for 6 h post-EMI exposure. A standard external cardiac defibrillator served as a positive control. Muscle and nerve tissue histology adjacent to the EMI contacts were examined. Post-EMI shock skeletal muscle function was evaluated by analyzing the compound muscle action potentials (CMAPs) of the rectus femoris muscle. Maximal energy cardiac defibrillator pulses resulted in rhabdomyolysis and marked elevation of CPK, LDH, and AST 6 h post-shock. EMI field pulses resulted in the animals developing transient acidosis. CMAP amplitudes decreased approximately 50% after EMI and recovered to near-normal levels within 6 h. Within 6 h post-EMI exposure, blood CPK was mildly increased, LDH was normal, and no arrhythmia was observed. Minimal rhabdomyolysis was produced by the EMI pulses. These results suggest that EMI exposure is unlikely to cause extremity rhabdomyolysis in normal individuals. Bioelectromagnetics. © 2020 Bioelectromagnetics Society.

In an early complain warning system, we encounter a common problem - the lack of angry emotions for training classification models. Moreover, the recognition of angry emotion is more important than that of no-anger emotion. Based on this, the main purpose of this paper is to train an optimal model which achieves a high recall above a lower bound and a maximum of F1 score. It is divided into three aspects: 1) A variant of F1 score (TF1 score) takes recall above a lower bound and F1 score into consideration; 2) A Single Emotion Deep Neural Network (SEDNN) and its training process are designed to find an optimal model with a maximum of TF1 score. 3) A performance comparison of different methods is conducted on IEMOCAP and Emo-DB database. Extensive experiments show that when a BCE loss function or a focal loss function is used, the training process can find a model with a recall above a high threshold and a maximum of F1 score. Especially, SEDNN with the focal loss function performs better than SEDNN with the BCE loss function.

Global reasoning plays a significant role in many computer vision tasks which need to capture long-distance relationships. However, most current studies on global reasoning focus on exploring the relationship between pixels and ignore the critical role of the regions. In this paper, we propose an novel approach that explores the relationship between regions which have richer semantics than pixels. Specifically, we design a region aggregation method that can gather regional features automatically into a uniform shape, and adjust theirs positions adaptively for better alignment. To achieve the best performance of global reasoning, we propose various relationship exploration methods and apply them on the regional features. Our region-based global reasoning module, named ReGr, is end-to-end and can be inserted into existing visual understanding models without extra supervision. To evaluate our approach, we apply ReGr to fine-grained classification and action recognition benchmark tasks, and the experimental results demonstrate the effectiveness of our approach.

Multisource exchangeability models (MEMs), a Bayesian approach for dynamically integrating information from multiple clinical trials, are a promising approach for gaining efficiency in randomized controlled trials. When the supplementary trials are considerably larger than the primary trial, care must be taken when integrating supplementary data to avoid overwhelming the primary trial. In this paper, we propose “capping priors,” which controls the extent of dynamic borrowing by placing an a priori cap on the effective supplemental sample size. We demonstrate the behavior of this technique via simulation, and apply our method to four randomized trials of very low nicotine content cigarettes.

In this paper, the smaller plastic scintillator (polystyrene as solvent) tiles are polymerized and shaped by two methods, i.e. injection molding technology and thermal pressing technology by domestic manufacturer, to be operated for more than 10 years in the hadron calorimeter of the Circular Electron Position Collider (CEPC) or other purposes. Other benefits include, researching aging problems, by adding 0.1% material antioxidants of 2,6-Di-tert-butyl-4-methylphenol (BHT) and without BHT, and according to Van't Hoff rule of equivalent chemical reaction rate of high temperature, the 10 years aging study is simulated in short term. A strobed coincidence setup is arranged to test the light yield of the tile to be measured with silicon photomultipliers (SiPMs) with ADC read out and a simple trigger system with small solid angle acceptance is adopted. The light yields of measured tiles without BHT decrease by about 26.6% (injection molding technology) and 43.3% (thermal polymerization technology) after ten-year equivalent aging time. Meanwhile, the light yields of the tiles with BHT almost did not change after ten-year equivalent aging time.

As a novel X-ray focusing technology, lobster eye micro-pore optics (MPO) feature both a wide observing field of view and true imaging capability, promising sky monitoring with significantly improved sensitivity and spatial resolution in soft X-rays. Since first proposed by Angel (1979), the optics have been extensively studied, developed and trialed over the past decades. In this Letter, we report on the first-light results from a flight experiment of the Lobster Eye Imager for Astronomy ($LEIA$), a pathfinder of the wide-field X-ray telescope of the Einstein Probe mission. The piggyback imager, launched in July 2022, has a mostly un-vignetted field of view of $18.6^\circ \times 18.6^\circ $. Its spatial resolution is in the range of 4$-$7 arcmin in FWHM and the focal spot effective area is 2$-$3 cm$^2$, both showing only mild fluctuations across the field of view. We present images of the Galactic center region, Sco X-1 and the diffuse Cygnus Loop nebular taken in snapshot observations over 0.5$-$4 keV. These are truly wide-field X-ray images of celestial bodies observed, for the first time, by a focusing imaging telescope. Initial analyses of the in-flight data show excellent agreement between the observed images and the on-ground calibration and simulations. The instrument and its characterization are briefly described, as well as the flight experiment. The results provide a solid basis for the development of the present and proposed wide-field X-ray missions using lobster eye MPO., Comment: 11 pages, 4 figures. Accepted for publication in Astrophysical Journal Letter

Background Our prospective, open-label, single-arm phase II study investigated the safety and efficacy of DCVAC/LuCa (dendritic cell vaccines for lung cancer) combined with standard carboplatin/pemetrexed in advanced non-squamous (nsq) non-small-cell lung cancer (NSCLC). Patients and methods Eligible patients had stage IV nsq NSCLC without oncogenic drivers and had not received prior systemic cancer therapy. Treatment consisted of carboplatin/pemetrexed for up to 6 cycles followed by 21 cycles of pemetrexed maintenance or until progression or intolerance. Non-progression patients after two cycles of chemotherapy started to receive DCVAC/LuCa subcutaneously (s.c.) on day 15 of cycle 3, and thereafter q3w (day 15 of chemotherapy cycles) for up to 15 doses. Dosing of DCVAC/LuCa s.c. varied among patients depending on the baseline number of leucocytes but remained constant for each single patient. Safety was assessed by adverse events (AEs), treatment-related adverse events (TRAEs), serious adverse events (SAEs), and adverse events of special interest (AESIs). Efficacy was measured by overall survival (OS), progression-free survival (PFS), time to progression (TTP), and objective response rate (ORR). Results Sixty-one patients were enrolled. In the safety population (n = 60), eight patients (13.33%) had grade 3 or greater TRAEs, and six patients (10.0%) showed SAEs which were not related to leukapheresis or DC vaccination. Six grade 1 AEs were considered to be related to leukapheresis. No AESIs or DCVAC/LuCa-induced AEs were observed. The 2-year survival rate in the modified intention-to-treat population (n = 44) was 52.57%. Median OS was not reached. Median PFS was 8.0 months, median TTP was 10.2 months, and the ORR was 31.82%. Conclusion In treatment-naïve stage IV nsq NSCLC patients without oncogenic drivers, the combination of carboplatin/pemetrexed and DCVAC/LuCa was well tolerated and showed promising efficacy. Therefore, a study to prove our immunotherapeutic concept in a randomized phase III trial is planned., Highlights • We investigated the safety and efficacy of DCVAC/LuCa combined with standard carboplatin/pemetrexed in nsq NSCLC. • The combination therapy showed a favorable tolerability profile in a selected Chinese population. • The 2-year survival rate in the modified intention-to-treat population (n = 44) was 52.57%.

This paper discusses an interactive white cane learning tool for children with visual impairment to familiarize themselves with the use of the white cane. A prototype, named CaneFX, was designed based on motion capture technology which utilizes an accelerometer to understand the motion of the white cane. A microcontroller was implemented to process the data obtained from the accelerometer. Three modes were implemented to detect different white cane motions which are sweeping, tapping and hitting motions and for each motion, a sound is associated with it. Experimental testing was conducted to find an optimal setup of the proposed prototype and evaluate its usability. The results show that the prototype is capable of detecting the motions in real-time with high accuracy on a rough floor surface.

OBJECTIVES Tuberculosis (TB) is the most common and fatal opportunistic co-infection among HIV-infected individuals. While TB-associated mortality predominantly occurs in the first 90 days after admission, such a correlation remains unclear in HIV/TB co-infected patients. Thus, we aimed to investigate the 90-day mortality and associated risk factors among HIV/TB co-infected patients in China. METHODS Adult patients with HIV and a newly confirmed TB diagnosis admitted to the Shanghai Public Health Clinical Center between September 2009 and August 2017 were enrolled. Clinical and laboratory characteristics, key treatments and outcomes were collected retrospectively. The associations between different factors and early mortality were analysed. RESULTS Of the 485 laboratory-confirmed HIV/TB patients [median (range) age = 39 (19-79) years], 413 (85.15%) were male. Diagnosis was confirmed by culture, pathology and acid-fast bacilli smear alone in 362 (74.6%), 6 (1.2%) and 117 (24.1%) patients, respectively. Multiple drug-/rifampin-resistant TB was detected in 21 (5.8%) of the 367 patients with a positive culture. Rifampin or rifabutin was administered to 402 (82.9%) patients. Additionally, 66 (13.6%) and 86 (17.7%) died within 90 days and 1 year of admission, respectively. Of the 64 TB-related deaths, 59 (92.2%) occurred within 90 days of admission. In Cox regression, central nervous system (CNS) TB [odds ratio (OR) = 2.49, 95% confidence interval (CI): 1.46-4.23, P

Standalone photovoltaic power system is one of the feasible solutions in rural electrification. However, due to the intermittent nature of solar energy and significant demand variations due to small load size, energy storage devices are needed to ensure power quality and reliability. Lead-acid battery bank has been widely used in such systems to act as the intermediate buffer to compensate the power mismatch between generation and demand, despite the poor lifetime characteristics and relatively short service life compared to other energy storage devices. Most charge controllers only provide basic battery charging functionality and prevent the battery bank from overcharging and deep-discharging. While other notable aging factors such as frequent charge-discharge transition, high C-rate and surge power are often neglected. This paper presents a smart hybrid energy storage plug-in module that aims to enhance the service life of Lead-acid battery in standalone photovoltaic-battery power systems by mitigating life-limiting factors such as current fluctuations and surge current. This module is designed as a plug-in module that can be adopted directly in existing infrastructure in installed standalone PV-battery power system. The effectiveness of the proposed plug-in module in mitigating battery stresses is simulated in Matlab Simulink with actual solar irradiance data and estimated load profile for a rural community in Sarawak Malaysia. The simulation results are validated experimentally with the scaled-down photovoltaic-battery power system. The effectiveness in mitigating battery stress is quantitatively evaluated with battery health cost function, while the reduction in overall battery cost is analyzed and discussed.

Model selection with cross validation (CV) is very popular in machine learning. However, CV with grid and other common search strategies cannot guarantee to find the model with minimum CV error, which is often the ultimate goal of model selection. Recently, various solution path algorithms have been proposed for several important learning algorithms including support vector classification, Lasso, and so on. However, they still do not guarantee to find the model with minimum CV error. In this paper, we first show that the solution paths produced by various algorithms have the property of piecewise linearity. Then, we prove that a large class of error (or loss) functions are piecewise constant, linear, or quadratic w.r.t. the regularization parameter, based on the solution path. Finally, we propose a new generalized error path algorithm (GEP), and prove that it will find the model with minimum CV error in a finite number of steps for the entire range of the regularization parameter. The experimental results on a variety of datasets not only confirm our theoretical findings, but also show that the best model with our GEP has better generalization error on the test data, compared to the grid search, manual search, and random search.

Large sparse datasets are very common in product recommendation applications. For the scalability reason, linear classifiers are preferred in classification tasks on such datasets. Our previous work Li et al. (2016) has studied how sparsity affects naive Bayes classification under typical data missing mechanisms. In this paper, we greatly expand our previous work by including all linear classifiers, and explore practical strategies to improve accuracy of large sparse data classification. Using real-world and synthetic experiments, we observe different learning curve behaviors under different missing mechanisms. We also study the theoretic reasons for all our observations. Our studies provide a practical guideline to determine if or when obtaining more data and/or obtaining missing values in the data is worthwhile or not. This can be very valuable in the recommendation system applications.

This paper introduces a novel global patch matching method that focuses on how to remove fronto-parallel bias and obtain continuous smooth surfaces with assuming that the scenes covered by stereos are piecewise continuous. Firstly, simple linear iterative cluster method (SLIC) is used to segment the base image into a series of patches. Then, a global energy function, which consists of a data term and a smoothness term, is built on the patches. The data term is the second-order Taylor expansion of correlation coefficients, and the smoothness term is built by combing connectivity constraints and the coplanarity constraints are combined to construct the smoothness term. Finally, the global energy function can be built by combining the data term and the smoothness term. We rewrite the global energy function in a quadratic matrix function, and use least square methods to obtain the optimal solution. Experiments on Adirondack stereo and Motorcycle stereo of Middlebury benchmark show that the proposed method can remove fronto-parallel bias effectively, and produce continuous smooth surfaces.

Machine learning, especially deep learning, has been successfully applied to a wide array of computer vision classification tasks in recent years. Infamous for requiring massive amounts of data to perform well at image classification problems, deep learning has so far been unable to solve Bongard problems (BPs), a set of abstract visual reasoning tasks invented in the 1960s. Each BP can be seen as a supervised learning task, with few training samples (6 for positive and 6 for negative), and often requiring highly abstract features to learn well. Automatically solving Bongard problems directly from images remains an ambitious goal, with very little machine learning literature on the topic. In this paper, we discuss several special properties of BPs as well as what it means to solve a BP. Making use of an expanded set of BP-like tasks to allow for a more careful evaluation of automated solvers, we develop and benchmark a deep learning based approach to solve these problems. To encourage work on this interesting problem, we also make freely available a dataset of over 200 BPs (https://github.com/XinyuYun/bongard-problems).

To advance understanding on how to engage readers, we advocate the novel task of automatic pull quote selection. Pull quotes are a component of articles specifically designed to catch the attention of readers with spans of text selected from the article and given more salient presentation. This task differs from related tasks such as summarization and clickbait identification by several aspects. We establish a spectrum of baseline approaches to the task, ranging from handcrafted features to a neural mixture-of-experts to cross-task models. By examining the contributions of individual features and embedding dimensions from these models, we uncover unexpected properties of pull quotes to help answer the important question of what engages readers. Human evaluation also supports the uniqueness of this task and the suitability of our selection models. The benefits of exploring this problem further are clear: pull quotes increase enjoyment and readability, shape reader perceptions, and facilitate learning. Code to reproduce this work is available at https://github.com/tannerbohn/AutomaticPullQuoteSelection.

Large antenna plays an important role in exploring outer space, which is a typical representative of complex systems. In this paper, the method of applying MBSE into the system design of large antenna is addressed, and the reasoning processes of requirement analysis, function decomposition and architecture mapping are represented by system data models. Requirement diagram of SysML is taken to exhibit requirement architecture of different relationships visually. Use case diagram of SysML builds a bridge between stakeholder needs and functional decomposition, and use case is decomposed level by level to sub-functions by activity diagram of SysML. The Block Definition Diagram (BDD) of SysML is generated form the mapping of function, and the Internal Block Diagram (IBD) is used to analyze the communication of different components. By virtue of MBSE, a front new pattern of schematic design is developed in the conceptual design of large antenna, which will be very helpful to elevate design quality and efficiency.

In the process of close-range photogrammetry, it is hard to stick markers directly on the surface of the object to be measured at times. Projecting optical markers is a good solution. However, there are some problems for the existing optical markers such as difficulty in automatic matching and splicing. Therefore, this paper proposes the design and detection algorithm of white-light markers. White-light marker is composed of collar coded targets and circular non-coded targets. Collar coded targets are designed in the middle of the white-light marker, which can enhance the brightness of the coded strip and improve the recognition accuracy. Circular non-coded targets are designed around the collar coded targets, which can realize matching of non-coded targets among multiple images. Based on Canny operator, feature edges are extracted from the images. According to the criterions of scale, form, etc., the edges of markers are screened out. Markers containing coded strip is decoded to determine the identity. Center of gravity method is used to make the center position of white-light makers come true. The designed white-light marker is applied to the deformation detection experiment of crane girder. Its deformation curve is basically similar with the curve of sticking markers, which verifies the feasibility of the deformation detection of crane girder based on white-light markers.

Extensive work on matrix factorization (MF) techniques have been done recently as they provide accurate rating prediction models in recommendation systems. Additional extensions, such as neighbour-aware models, have been shown to improve rating prediction further. However, these models often suffer from a long computation time. In this paper, we propose a novel method that applies clustering algorithms to the latent vectors of users and items. Our method can capture the common interests between the cluster of users and the cluster of items in a latent space. A matrix factorization technique is then applied to this cluster-level rating matrix to predict the future cluster-level interests. We then aggregate the traditional user-item rating predictions with our cluster-level rating predictions to improve the rating prediction accuracy. Our method is a general "wrapper" that can be applied to all collaborative filtering methods. In our experiments, we show that our new approach, when applied to a variety of existing matrix factorization techniques, improves their rating predictions and also results in better rating predictions for cold-start users. Above all, in this paper we show that better quality and more quantity of these clusters achieve a better rating prediction accuracy.

A phenomenological-based micromechanical method, comprising the coupling of the matrix constitutive model and the cohesive zone (CZM) model at fiber-matrix interfaces, is presented to investigate the mechanical behaviors of unidirectional (UD) fiber-reinforced polymer (FRP) composites subjected to transverse tension and compression. The key point of this method is the establishment of a novel elasto-plastic damage constitutive model using a fracture plane based yield criterion. Both plastic deformation and progressive failure procedure is incorporated in the implicit simulation. Special focus is given to the determination of the input values of constituent material properties, suggesting that directly using the macroscopic matrix properties to characterize the mechanical responses at the microscale level may bring large discrepancies in homogenized stress-strain responses of UD composites. A parametrical study is also carried out to calibrate cohesive parameters. The numerical results, in good agreement with experimental measurements, clearly reveal the damage scenarios and the dominant failure mechanisms. It can be concluded that the debonding of the fiber-matrix interfaces is responsible for the ultimate transverse tensile strength, while compression failure is governed by two possible modes, i.e. matrix compressive failure if matrix failure firstly occurs, and matrix tensile and interface debonding failure when debonding initiates first. Besides, the distribution of fracture plane angles in matrix is examined to further validate the predictive capacity of the proposed model.