Portrait Segmentation Using Ensemble of Heterogeneous Deep-Learning Models

Image segmentation plays a central role in a broad range of applications, such as medical image analysis, autonomous vehicles, video surveillance and augmented reality. Portrait segmentation, which is a subset of semantic image segmentation, is widely used as a preprocessing step in multiple applications such as security systems, entertainment applications, video conferences, etc. A substantial amount of deep learning-based portrait segmentation approaches have been developed, since the performance and accuracy of semantic image segmentation have improved significantly due to the recent introduction of deep learning technology. However, these approaches are limited to a single portrait segmentation model. In this paper, we propose a novel approach using an ensemble method by combining multiple heterogeneous deep-learning based portrait segmentation models to improve the segmentation performance. The Two-Models ensemble and Three-Models ensemble, using a simple soft voting method and weighted soft voting method, were experimented. Intersection over Union (IoU) metric, IoU standard deviation and false prediction rate were used to evaluate the performance. Cost efficiency was calculated to analyze the efficiency of segmentation. The experiment results show that the proposed ensemble approach can perform with higher accuracy and lower errors than single deep-learning-based portrait segmentation models. The results also show that the ensemble of deep-learning models typically increases the use of memory and computing power, although it also shows that the ensemble of deep-learning models can perform more efficiently than a single model with higher accuracy using less memory and less computing power.     

Ensemble Neuroevolution-Based Approach for Multivariate Time Series Anomaly Detection

Multivariate time series anomaly detection is a widespread problem in the field of failure prevention. Fast prevention means lower repair costs and losses. The amount of sensors in novel industry systems makes the anomaly detection process quite difficult for humans. Algorithms that automate the process of detecting anomalies are crucial in modern failure prevention systems. Therefore, many machine learning models have been designed to address this problem. Mostly, they are autoencoder-based architectures with some generative adversarial elements. This work shows a framework that incorporates neuroevolution methods to boost the anomaly detection scores of new and already known models. The presented approach adapts evolution strategies for evolving an ensemble model, in which every single model works on a subgroup of data sensors. The next goal of neuroevolution is to optimize the architecture and hyperparameters such as the window size, the number of layers, and the layer depths. The proposed framework shows that it is possible to boost most anomaly detection deep learning models in a reasonable time and a fully automated mode. We ran tests on the SWAT and WADI datasets. To the best of our knowledge, this is the first approach in which an ensemble deep learning anomaly detection model is built in a fully automatic way using a neuroevolution strategy.     

Enhancing convolutional neural network scheme for rheumatoid arthritis grading with limited clinical data

The gray-scale ultrasound(US) imaging method is usually used to assess synovitis in rheumatoid arthritis(RA) in clinical practice. This four-grade scoring system depends highly on the sonographer's experience and has relatively lower validity compared with quantitative indexes. However, the training of a qualified sonographer is expensive and timeconsuming while few studies focused on automatic RA grading methods. The purpose of this study is to propose an automatic RA grading method using deep convolutional neural networks(DCNN) to assist clinical assessment. Gray-scale ultrasound images of finger joints are taken as inputs while the output is the corresponding RA grading results. Firstly,we performed the auto-localization of synovium in the RA image and obtained a high precision in localization. In order to make up for the lack of a large annotated training dataset, we performed data augmentation to increase the number of training samples. Motivated by the approach of transfer learning, we pre-trained the GoogLeNet on ImageNet as a feature extractor and then fine-tuned it on our own dataset. The detection results showed an average precision exceeding 90%. In the experiment of grading RA severity, the four-grade classification accuracy exceeded 90% while the binary classification accuracies exceeded 95%. The results demonstrate that our proposed method achieves performances comparable to RA experts in multi-class classification. The promising results of our proposed DCNN-based RA grading method can have the ability to provide an objective and accurate reference to assist RA diagnosis and the training of sonographers.     

Characterization of pure high-grade DCIS on magnetic resonance imaging using the evolving breast MR lexicon terminology: can it be differentiated from pure invasive disease?

Magnetic resonance imaging (MRI) is now a recognized method of imaging the breast. Unfortunately, there is lack of standardization in the MRI terminology used to characterize the appearance of breast lesions. Moreover, cases of mixed histologies are often imaged. We retrospectively identified cases of pure high-grade ductal carcinoma in situ (DCIS) using the recently introduced breast MRI lexicon and characterized the lesions in order to try and identify features that might distinguish high-grade DCIS from invasive disease. Five-year review of our institution's database revealed 637 patients underwent gadolinium-enhanced breast MRI examination. Twenty patients had histologically proven pure high-grade DCIS. After excluding patients with previous chemotherapy or inadequate MRI examination, 13 patients were analyzed and compared to the 13 most recent cases of pure invasive breast carcinoma. The morphological and dynamic features were then compared. High-grade DCIS cases were significantly more likely to show focal branching pattern (P=.03) and to have an irregular contour (P=.03), compared with invasive disease. Although of marginal statistical significance, DCIS lesions are more likely to have a lower morphological score than invasive carcinoma (P=.06), whilst the latter is more likely to show ring enhancement (P=.07). Use of breast MRI for staging at our institution shows that pure DCIS and pure invasive cancers are both rare entities. Despite the relatively limited numbers, we identified features that would help to differentiate high-grade DCIS from invasive carcinoma on MRI.     

Multimodal Fake-News Recognition Using Ensemble of Deep Learners

Social networks have drastically changed how people obtain information. News in social networks is accompanied by images and videos and thus receives more attention from readers as opposed to traditional sources. Unfortunately, fake-news publishers often misuse these advantages to spread false information rapidly. Therefore, the early detection of fake news is crucial. The best way to address this issue is to design an automatic detector based on fake-news content. Thus far, many fake-news recognition systems, including both traditional machine learning and deep learning models, have been proposed. Given that manual feature-extraction methods are very time-consuming, deep learning methods are the preferred tools. This study aimed to enhance the performance of existing approaches by utilizing an ensemble of deep learners based on attention mechanisms. To a great extent, the success of an ensemble model depends on the variety of its learners. To this end, we propose a novel loss function that enforces each learner to attend to different parts of news content on the one hand and obtain good classification accuracy on the other hand. Also, the learners are built on a common deep-feature extractor and only differ in their attention modules. As a result, the number of parameters is reduced efficiently and the overfitting problem is addressed. We conducted several experiments on some widely used fake-news detection datasets. The results confirm that the proposed method consistently surpasses the existing peer methods.     

A Maximum Entropy Method for Particle Filtering

Standard ensemble or particle filtering schemes do not properly represent states of low priori probability when the number of available samples is too small, as is often the case in practical applications. We introduce here a set of parametric resampling methods to solve this problem. Motivated by a general H-theorem for relative entropy, we construct parametric models for the filter distributions as maximum-entropy/minimum-information models consistent with moments of the particle ensemble. When the prior distributions are modeled as mixtures of Gaussians, our method naturally generalizes the ensemble Kalman filter to systems with highly non-Gaussian statistics. We apply the new particle filters presented here to two simple test cases: a one-dimensional diffusion process in a double-well potential and the three-dimensional chaotic dynamical system of Lorenz.     

Derivation of Mayer Series from Canonical Ensemble

Mayer derived the Mayer series from both the canonical ensemble and the grand canonical ensemble by use of the cluster expansion method. In 2002, we conjectured a recursion formula of the canonical partition function of a fluid(X.Z. Wang, Phys. Rev. E66(2002) 056102). In this paper we give a proof for this formula by developing an appropriate expansion of the integrand of the canonical partition function. We further derive the Mayer series solely from the canonical ensemble by use of this recursion formula.     

Financial Fraud Detection and Prediction in Listed Companies Using SMOTE and Machine Learning Algorithms

This paper proposes a new method that can identify and predict financial fraud among listed companies based on machine learning. We collected 18,060 transactions and 363 indicators of finance, including 362 financial variables and a class variable. Then, we eliminated 9 indicators which were not related to financial fraud and processed the missing values. After that, we extracted 13 indicators from 353 indicators which have a big impact on financial fraud based on multiple feature selection models and the frequency of occurrence of features in all algorithms. Then, we established five single classification models and three ensemble models for the prediction of financial fraud records of listed companies, including LR, RF, XGBOOST, SVM, and DT and ensemble models with a voting classifier. Finally, we chose the optimal single model from five machine learning algorithms and the best ensemble model among all hybrid models. In choosing the model parameter, optimal parameters were selected by using the grid search method and comparing several evaluation metrics of models. The results determined the accuracy of the optimal single model to be in a range from 97% to 99%, and that of the ensemble models as higher than 99%. This shows that the optimal ensemble model performs well and can efficiently predict and detect fraudulent activity of companies. Thus, a hybrid model which combines a logistic regression model with an XGBOOST model is the best among all models. In the future, it will not only be able to predict fraudulent behavior in company management but also reduce the burden of doing so.     

Fluctuation relations for heat exchange in the generalized Gibbs ensemble

In this work, we investigate the heat exchange between two quantum systems whose initial equilibrium states are described by the generalized Gibbs ensemble. First, we generalize the fluctuation relations for heat exchange discovered by Jarzynski and Wójcik to quantum systems prepared in the equilibrium states described by the generalized Gibbs ensemble at various generalized temperatures. Secondly, we extend the connections between heat exchange and the Rényi divergences to quantum systems under generic initial conditions. These relations are applicable for quantum systems with conserved quantities and universally valid for quantum systems in the integrable and chaotic regimes.     

基于Lorenz模型的集合预报与单一预报的比较研究

根据非线性局部Lyapunov向量方法和增长模繁殖方法,选取Lorenz63模型和Lorenz96模型的不同状态为例,对集合预报与单一预报的预报技巧开展了对比研究.结果表明:与单一预报比较,集合预报的均方根误差和型异常相关有明显改善,随预报时间推移,改善效果越显著,且集合平均优于单一预报的实验个例数逐渐增多.就概率分布(f)而言,单一预报状态的f与真实状态基本一致,不随时间变化;而集合平均预报状态的f则随时间呈现出值域变窄、峰值变大的特点.表明随预报时间的延长,单一预报状态为混沌吸引子上的随机状态,而集合平均预报状态为吸引子子集上的随机状态,这可能是集合平均误差小于单一预报的原因.     

Quantum leakage of collective excitations of atomic ensemble induced by spatial motion

We generalize the conception of quantum leakage for the atomic collective excitation states. By making use of the atomic coherence state approach, we study the influence of the atomic spatial motion on the symmetric collective states of 2-level atomic ensemble due to inhomogeneous coupling. In the macroscopic limit, we analyze the quantum decoherence of the collective atomic state by calculating the quantum leakage for a very large ensemble at a finite temperature. Our investigations show that the fidelity of the atomic system will not be good in the case of atom numberN→∞. Therefore, quantum leakage is an inevitable problem in using the atomic ensemble as a quantum information memory. The detailed calculations shed theoretical light on quantum processing using atomic ensemble collective qubit.     

Finite-size scaling in a microcanonical ensemble

The finite-size scaling technique is extended to a microcanonical ensemble. As an application, equilibrium magnetic properties of anL×L square lattice Ising model are computed using the microcanonical ensemble simulation technique of Creutz, and the results are analyzed using the microcanonical ensemble finite-size scaling. The computations were done on the multitransputer system of the Condensed Matter Theory Group at the University of Mainz.     

Sound speed correction in ultrasound imaging

A constant sound speed of 1.54 mm/micros is generally used by ultrasound imaging systems for delay and timing. However, the body's sound speed in-homogeneity can lead to defocusing and increased clutter. To provide an improvement using standard transducers, the sound speed used in delay and timing was computed using different sound speeds. We observed improvement in lateral resolution and clutter in phantom, OB, abdominal, and breast imaging. In OB and abdominal imaging using a 4 MHz curved array, 1.48 mm/micros provided higher image quality in many situations. In breast with an 8 MHz linear array, 1.44 mm/micros provided better images in some cases. To provide an automated way to determine and adjust the sound speed used by the imaging system, an algorithm was developed that determines the sound speed that produces the best overall lateral image quality by analyzing the spatial frequency content in a single B-mode frame of channel data using images reconstructed using various trial sound speeds. The metric produced correlates well with the observed best lateral image quality.     

Isospin dependence of nuclear multifragmentation in statistical model

The evolution of nuclear disintegration mechanisms with increasing excitation energy, from com- pound nucleus to multifragmentation, has been studied by using the Statistical Multifragmentation Model (SMM) within a micro-canonical ensemble. We discuss the observable characteristics as functions of excitation energy in multifragmentation, concentrating on the isospin dependence of the model in its decaying mechanism and break-up fragment configuration by comparing the A<,0> = 200, Z<,0> = 78 and A<,0> = 200, Z<,0> = 100 systems. The calculations indicate that the neutron-rich system (Z<,0> = 78) translates to a fission-like process from evaporation later than the symmetric nucleus at a lower excitation energy, but gets a larger average multiplicity as the excitation energy increases above 1.0 MeV/u.     

利用扩展系综法计算水的自由能

利用扩展系综法得到了正则系综下水的TIP4P模型的自由能值为-21．485±0．035kJ/mol， 并与其他方法所得的结果作了比较.提出了选择该方法中关键参数(平衡因子)的有效方法， 并讨论了该方法的可移植性. 关键词： 自由能 TIP4P水模型 扩展系综 分子动力学模拟 水分子团     

Determination of the vapor–liquid transition of square-well particles using a novel generalized-canonical-ensemble-based method

The square-well(SW) potential is one of the simplest pair potential models and its phase behavior has been clearly revealed, therefore it has become a benchmark for checking new theories or numerical methods. We introduce the generalized canonical ensemble(GCE) into the isobaric replica exchange Monte Carlo(REMC) algorithm to form a novel isobaric GCE-REMC method, and apply it to the study of vapor–liquid transition of SW particles. It is validated that this method can reproduce the vapor–liquid diagram of SW particles by comparing the estimated vapor–liquid binodals and the critical point with those from the literature. The notable advantage of this method is that the unstable vapor–liquid coexisting states,which cannot be detected using conventional sampling techniques, are accessed with a high sampling efficiency. Besides,the isobaric GCE-REMC method can visit all the possible states, including stable, metastable or unstable states during the phase transition over a wide pressure range, providing an effective pathway to understand complex phase transitions during the nucleation or crystallization process in physical or biological systems.     

Accurate measurement of wall skin friction by single-pixel ensemble correlation

The present work deals with accurately estimating wall-skin friction from near-wall mean velocity by means of PIV measurement.The estimation accuracy relies on the spatial resolution and the precision of the resolved velocity profile inside the viscous sublayer,which is a big challenge for conventional window-based correlation method(K?hler C J,et al.Exp Fluids,2012,52:1641–1656).With the help of single-pixel ensemble correlation,the ensemble-averaged velocity vector can be resolved at significant spatial resolution,thus improving the measurement accuracy.To demonstrate the feasibility of this single-pixel ensemble correlation method,we first study the velocity estimation precision in a case of steady near-wall flow.Synthetic particle images are used to investigate the effect of different image parameters.It is found that the velocity RMS-uncertainty level of the single-pixel ensemble correlation method can be equivalent to the conventional window correlation method once the effective particle number used for the ensemble correlation is large enough.Furthermore,a canonical turbulent boundary layer is synthetically simulated based on velocity statistics resolved by previous Direct Numerical Simulation(DNS)work(Schlatter P,et al.J Fluid Mech,2010,659:116–126).The relative error of wall skin friction coefficient is shown to be one-order smaller than that of the window correlation method.And the optimization strategy to further minimize the measurement uncertainty is discussed in the last part.