Search

The material constriction is one of the important factors that influence the forming accuracy of selective laser sintering (SLS). Currently, in order to reduce the shrinkage and improve the quality of products, the optimal combination of machining process parameters is mainly determined by numerous experiments. This often takes valuable time and costs a lot, but the results are mediocre. With the development of intelligent optimization algorithms, they are applied in various disciplines for solving complex problems. Hence, for reducing the shrinkage of parts and overcoming the limitation in the optimization of the process parameters, this paper proposes a novel hybrid improved Hunger Games Search algorithm (HGS) with extreme learning machine (ELM) model for predicting the shrinkage of parts. Firstly, the orthogonal experiments were conducted based on the five key process parameters, the obtained parts datasets were divided into the training set and test set. Secondly, the Cube mapping and refracted opposition-based learning strategies are adopted to increase the convergence speed and solution accuracy of HGS. In addition, the regression prediction model was constructed with the improved HGS(IHGS) and ELM, and this model is trained using the training set. Finally, the test set is used to evaluate the trained model and find the optimal combination of process parameters with the lowest shrinkage of parts. The experimental results suggest that the IHGS-ELM model proposed in this study has high forecasting precision, with the R2 and RMSE are only 0.9124 and 0.2433, respectively. This model can guide the laser sintering process of polyether sulfone (PES) powder.

Relation triple extraction is a combination of named entity recognition and relation prediction. Early works ignore the problem of data overlap when extracting triples, resulting in poor extraction performance. Subsequent works improve the capability of the model to extract overlapping triples through generative and extractive methods. These works achieve considerable performance but still suffer from some defects, such as poor extraction capability for individual triplets and inappropriate spatial distribution of the data. To solve the above problems, we perform sequence-to-matrix transformation and propose the NEDORT model. NEDORT predicts all subjects in the sentence and then completes the extraction of relation–object pairs. There are overlapping parts between relation–object pairs, so we conduct the conversion of sequence to matrix. We design the Differential Amplified Multi-head Attention method to extract subjects. This method highlights the locations of entities and captures sequence features from multiple dimensions. When performing the extraction of relation–object pairs, we fuse subject and sequence information through the Biaffine method and generate relation–sequence matrices. In addition, we design a multi-layer U-Net network to optimize the matrix representation and improve the extraction performance of the model. Experimental results on two public datasets show that our model outperforms other baseline models on triples of all categories

Human pose estimation from video sequences has become a hot research topic in the domain of robotics and computer vision. However, existing three-dimensional (3D) pose estimation methods usually analyze individual frames, which have a low accuracy due to various human movement speed, limiting its practical application. In this paper, we propose a method for estimating 3D pose and calculating similarity from Tai Chi video sequences based on Seq2Seq network. Specifically, using 2D joint point coordinate sequence of the original image as input, our method constructs an encoder and a decoder to build a Seq2Seq network. Our method introduces an attention mechanism for weighing the input data to obtain an intermediate vector and decode it to estimate the 3D joint point sequence. Afterwards, using a template video and a target video as input, our method calculates the cost of passing through each point within the constraints to construct a cost matrix for video similarity. With the cost matrix, our method can determine the optimal path and use the correspondence of the video sequence to calculate the image similarity of the corresponding frame. The experimental data show that the proposed method can effectively improve the accuracy of 3D pose estimation, and increase the speed for video similarity calculation.

Machine learning is one of the most potential ways to realize the function prediction of the incremental large-scale G-protein-coupled receptors (GPCR). Prior research reveals that the key to determining the overall classification accuracy of GPCR is extracting valuable features and filtering out redundancy. To achieve a more efficient classification model, we put the feature synonym problem into consideration and create a new method based on functional word clustering and integration. Through evaluating the evolution correlation between features using the transition scores in mature molecular substitution matrices, candidate features are clustered into synonym groups. Each group of the clustered features is then integrated and represented by a unique key functional word. These retained key functional words are used to form a feature knowledge base. The original GPCR sequences are then transferred into feature vectors based on a feature re-extraction strategy according to the features in the knowledge base before the training and testing stage. We create multiple machine learning models based on Naïve Bayesian (NB), random forest (RF), support vector machine (SVM), and multi-layer perceptron (MLP) algorithms. The established model is applied to classify two public data sets containing 8354 and 12,731 GPCRs, respectively. These models achieve significant performance in almost all evaluation criteria in comparison with state-of-the art. This work demonstrated the potential of the novel feature extraction strategy and provided an effective theoretical design for the hierarchical classification of GPCRs.

The performance of deep neural networks is heavily dependent on its architecture and various neural architecture search strategies have been developed for automated network architecture design. Recently, evolutionary neural architecture search (EvoNAS) has received increasing attention due to the attractive global optimization capability of evolutionary algorithms. However, EvoNAS suffers from extremely high computational costs because a large number of performance evaluations are usually required in evolutionary optimization, and training deep neural networks is itself computationally very expensive. To address this issue, this article proposes a computationally efficient framework for the evolutionary search of convolutional networks based on a directed acyclic graph, in which parents are randomly sampled and trained on each mini-batch of training data. In addition, a node inheritance strategy is adopted so that the fitness of all offspring individuals can be evaluated without training them. Finally, we encode a channel attention mechanism in the search space to enhance the feature processing capability of the evolved neural networks. We evaluate the proposed algorithm on the widely used datasets, in comparison with 30 state-of-the-art peer algorithms. Our experimental results show that the proposed algorithm is not only computationally much more efficient but also highly competitive in learning performance.

Modern compilers provide a huge number of optional compilation optimization options. It is necessary to select the appropriate compilation optimization options for different programs or applications. To mitigate this problem, machine learning is widely used as an efficient technology. How to ensure the integrity and effectiveness of program information is the key to problem mitigation. In addition, when selecting the best compilation optimization option, the optimization goals are often execution speed, code size, and CPU consumption. There is not much research on program reliability. This paper proposes a Gate Graph Attention Neural Network (GGANN)-based compilation optimization option selection model. The data flow and function-call information are integrated into the abstract syntax tree as the program graph-based features. We extend the deep neural network based on GGANN and build a learning model that learns the heuristics method for program reliability. The experiment is performed under the Clang compiler framework. Compared with the traditional machine learning method, our model improves the average accuracy by 5–11% in the optimization option selection for program reliability. At the same time, experiments show that our model has strong scalability.

As an important positioning source of indoor positioning technology, Wi-Fi signals have attracted the attention of researchers for a long time. Fingerprint positioning can solve the problems caused by non-line-of-sight propagation and multipath effects. To improve the accuracy of Wi-Fi indoor positioning, this study proposes an indoor positioning algorithm based on fine-grained channel state information (CSI) and convolutional neural network (CNN). CSI is a kind of observable measurement that better describes the nature of Wi-Fi signal propagation than received signal strength indication. This method uses the subcarrier amplitude and phase difference information extracted from CSI data to establish fingerprints. The clustering method is used to analyse the number of clusters of fingerprint data, and the fingerprint database is divided into two sub-databases according to the threshold. CNNs with the same network structure are used to train the two kinds of fingerprint sub-databases. In the positioning stage, the sub-database to which the data to be measured belongs is determined according to the calibration algorithm, and the corresponding CNN model is used to estimate the position. Experiments are performed in a typical indoor environment. Compared with existing fingerprint-based positioning methods, this method has higher positioning accuracy.

Millimetre wave radar is an emerging technology that can monitor vital signs without contact. This unique feature is very suitable for some particular situations, such as burn patient monitoring. C...

Likelihood probability based phylogenetic analysis approaches have contributed to impressive advances in minimizing the variance of estimating the evolutionary parameters. However, their actual applications are greatly limited due to the very time-consuming calculations of Conditional Likelihood Probabilities (CLPs). Accurately and quickly obtaining the likelihoods of massive tree samples can facilitate phylogenetic analysis process. Inspired by recent advance of machine learning techniques that greatly improve the performance of many related prediction tasks, this study proposes a Random Forest (RF) based learning and prediction approach, called NeoPLE. The approach initially learns the deep neighbor information between nodes from the topology representations of evolution trees, integrates likelihood information from these trees, and trains a non-linear prediction model. Instead of having to depend on the recursive calculations of the CLPs of tree nodes, NeoPLE transfers the process to a prediction by the trained model, thus the likelihood estimates become irrelevant with the calculations of CLPs. In terms of performance improvement, speedup factors ranging from 2.1 to 3.5X can be achieved on the analysis of realistic data sets. Moreover, NeoPLE is very suitable to handle the data sets having relatively large number of alignment sites, the factor of up to 27.5X can be achieved on the analysis of simulated data sets. In addition, NeoPLE is robust against a wide range of choices of evolutionary models and is ready to integrate in more phylogenetic inference tools. This study fills in the gaps of phylogenetic analysis using a machine learning approach in feature representation and likelihood prediction of evolution trees, which has not been reported in literatures.

Healthy lifestyle trends are getting more prominent around the world. There are numerous numbers of marathon running race events that have been held and inspired interest among peoples of different ages, genders and countries. Such diversified truths increase more difficulties to comprehending large numbers of marathon images, since such process is often done manually. Therefore, a new approach for racing bib number (RBN) localization and recognition for marathon running races using deep learning is proposed in this paper. Previously, all RBN application systems have been developed by using image processing techniques only, which limits the performance achieved. There are two phases in the proposed system that are phase 1: RBN detection and phase 2: RBN recognition. In phase 1, You Only Look Once version 3 (YOLOv3) which consists of a single convolutional network is used to predict the runner and RBN by multiple bounding boxes and class probabilities of those boxes.YOLOv3 is a new classifier network that outperforms other state-of-art networks. In phase 2, Convolutional Recurrent Neural Network (CRNN) is used to generate a label sequence for each input image and then select the label sequence that has the highest probability. CRNN can be straight trained from sequence labels such as words without any annotation of characters. Therefore, CRNN recognizes the contents of RBN detected. The experimental results based on mean average precision (mAP) and edit distance have been analyzed. The developed system is suitable for marathon or distance running race events and automates the localization and recognition of racers, thereby increasing efficiency in event control and monitoring as well as post-processing the event data

Graph Convolutional Networks (GCNs) are powerful representation learning methods for non-Euclidean data. Compared with the Euclidean data, labeling the non-Euclidean data is more expensive. Meanwhile, most existing GCNs only utilize few labeled data but ignore most of the unlabeled data. To address this issue, we design a novel end-to-end Iterative Feature Clustering Graph Convolutional Networks (IFC-GCN) that enhances the standard GCN with an Iterative Feature Clustering (IFC) module. The proposed IFC module constrains node features iteratively based on the predicted pseudo labels and feature clustering. Further, we design an EM-like framework for IFC-GCN training, which improves the network performance by rectifying the pseudo labels and the node features alternately. Theoretical analysis and experimental results show that our proposed IFC module can effectively modify the node features. Experimental results on public datasets demonstrate that IFC-GCN outperforms state-of-the-art methods on the semi-supervised node classification task.

The recent success of Transformer has provided a new direction to various visual understanding tasks, including video-based facial expression recognition (FER). By modeling visual relations effectively, Transformer has shown its power for describing complicated patterns. However, Transformer still performs unsatisfactorily to notice subtle facial expression movements, because the expression movements of many videos can be too small to extract meaningful spatial-temporal relations and achieve robust performance. To this end, we propose to decompose each video into a series of expression snippets, each of which contains a small number of facial movements, and attempt to augment the Transformer's ability for modeling intra-snippet and inter-snippet visual relations, respectively, obtaining the Expression snippet Transformer (EST). In particular, for intra-snippet modeling, we devise an attention-augmented snippet feature extractor (AA-SFE) to enhance the encoding of subtle facial movements of each snippet by gradually attending to more salient information. In addition, for inter-snippet modeling, we introduce a shuffled snippet order prediction (SSOP) head and a corresponding loss to improve the modeling of subtle motion changes across subsequent snippets by training the Transformer to identify shuffled snippet orders. Extensive experiments on four challenging datasets (i.e., BU-3DFE, MMI, AFEW, and DFEW) demonstrate that our EST is superior to other CNN-based methods, obtaining state-of-the-art performance.

In the study, our group choose a set of quality of red wine as data set. To get a more accurate result, we turn the quality into binary classification. And we try to build models to predict the quality of red wine based on machine learning algorithms, including Decision Tree, Boosting, Classification and regression tree and Random Forest. Among them, CART and Random Forest both get a high accuracy. A binary tree is built with CART and feature importance is analyzed. Meanwhile, we try to combine logistic algorithm with Random Forest and compare the accuracy of different models. In this way, it's found that there is a way to improve the accuracy of these models.

In recent years, deep convolutional neural networks (CNNs) have been widely used for hyperspectral image (HSI) classification. Besides, ensemble learning is a useful way to enhance the classification performance. Therefore, in this study, a new method titled Boosting-CNN is proposed for HSI classification, which fully explored the advantages of deep CNN and ensemble learning. Specifically, several deep CNNs are well-designed to classify HSI. The samples are misclassified in a CNN have more weighs in the following CNN through adaptive boosting. The final classification result is obtained by weighted voting of several CNNs. For HSI classification issue, the number of samples in different classes varies greatly, however, traditional classification methods cannot handle this issue well. In order to address imbalance training samples in HSI classification, soft class balanced loss is proposed to mitigate the influence of imbalance training samples. Experimental results on two popular hyperspectral datasets (i.e., Salinas and Pavia University) show that the proposed method obtain better classification accuracy compared to comparison methods.

Wildfires may cause great casualties and heavy wildfires are becoming more and more frequently all over the world in recent years. However, due to the environmental limitation, high manual-dependent operation is often impractical with other limits. In this paper, a transfer learning neural network based on long short term memory (LSTM) was used to detect wildfire based on Himawari-8. The real time dynamic threshold value detection for cloud mask based on the modified Otsu algorithm was used to fast and accurately remove cloud areas where wildfire detection is failed due to signal blocking. Then, the experiments were conducted with LSTM and other models. The experimental results showed that our method was positive for wildfire detection.

Jamming is a big threat to the survival of a radar system. Therefore, the recognition of radar jamming signal type is a part of radar countermeasure. Recently, convolutional neural networks (CNNs) have shown their effectiveness in radar signal processing, including jamming signal recognition. However, most of existing CNN methods do not regard radar jamming as a complex value signal. In this study, a complex-valued CNN (CV-CNN) is investigated to fully explore the inherent characteristics of a radar jamming signal, and we find that we can obtain better recognition accuracy using this method compared with a real-valued CNN (RV-CNN). CV-CNNs contain more parameters, which need more inference time. To reduce the parameter redundancy and speed up the recognition time, a fast CV-CNN (F-CV-CNN), which is based on pruning, is proposed for radar jamming signal fast recognition. The experimental results show that the CV-CNN and F-CV-CNN methods obtain good recognition performance in terms of accuracy and speed. The proposed methods open a new window for future research, which shows a huge potential of CV-CNN-based methods for radar signal processing.

Face verification has come into increasing focus in various applications including the European Entry/Exit System, which integrates face recognition mechanisms. At the same time, the rapid advancement of biometric authentication requires extensive performance tests in order to inhibit the discriminatory treatment of travellers due to their demographic background. However, the use of face images collected as part of border controls is restricted by the European General Data Protection Law to be processed for no other reason than its original purpose. Therefore, this paper investigates the suitability of synthetic face images generated with StyleGAN and StyleGAN2 to compensate for the urgent lack of publicly available large-scale test data. Specifically, two deep learning-based (SER-FIQ, FaceQnet v1) and one standard-based (ISO/IEC TR 29794-5) face image quality assessment algorithm is utilized to compare the applicability of synthetic face images compared to real face images extracted from the FRGC dataset. Finally, based on the analysis of impostor score distributions and utility score distributions, our experiments reveal negligible differences between StyleGAN vs. StyleGAN2, and further also minor discrepancies compared to real face images.

In this paper, a parallel model adaptive Kalman filtering algorithm is presented for multiple sensors estimation fusion when the measurement noise statistics are uncertain. As a typical adaptive filtering algorithm, the multiple model adaptive estimation tries to reduce the dependency of the filter on the noise parameters. It utilizes multiple models with different noise levels to estimate the state and combines the model-dependent estimates with model probability. However, with the increase in the number of active sensors, a large number of models are required to cover the entire range of the possible noise parameter values, which can become computationally infeasible. The main goal of this work is to incorporate the noise statistic estimator in the framework of the multiple model adaptive estimation, such that only two models are required for each sensor, which significantly reduce the complexity of the estimator. The advantage of the presented algorithm to deal with the model uncertainty is studied analytically. The high performance of the parallel model adaptive Kalman filtering for autonomous satellite navigation using inter-satellite line-of-sight measurements is illustrated in comparison with a robust Kalman filter, an intrinsically Bayesian robust Kalman filter, and the traditional multiple model adaptive estimation.

The purpose of this study is to achieve an accurate classification of medical images of pneumonia using transfer learning. Pulmonary radiographs can provide physicians with an excellent aid in the diagnosis of pneumonia, and this paper collects images from patients with common bacterial pneumonia, COVID-19 disease, and normal incidents pulmonary x-rays are used as a dataset for transfer learning. The deep convolutional neural network framework proposed in recent years can be used to extract features from the dataset and classify them to obtain models with high accuracy. The transfer learning-based approach can not only reduce the dependence of the network model on the dataset but also significantly reduce the time and hardware costs, and the trained models usually produce significant results as well. A total of 4,500 x-ray images are collected in this paper, including 1,500 confirmed images of confirmed COVID-19 disease, 1,500 images of confirmed bacterial pneumonia, and 1,500 images of normal conditions. This paper, VGG19, ResNet-50, and GoogLeNet are used as the basis for transfer learning. The results show that using transfer learning can effectively extract advanced features of pneumonia and construct a pneumonia medical image recognizer, which is trained to recognize the models. The accuracy rates is 93.67%, 91.73%, and 90.80%, respectively, and the model shows excellent generalization ability. According to the results, this study can assist medical professionals in the diagnosis of pneumonia condition and effectively shorten the diagnosis time.

Aiming at the problem of weak signal and a lot of noise in non-contact vital signs detection, a band-pass filter is designed to effectively extract the breathing signal and heart signal, and the heart signal is classified by LSTM network to identify the patients with atrial fibrillation (AFib) disease. Firstly, the vital signs obtained by millimeter wave radar is processed by range FFT, then perform phase extraction and phase unwrapping on the range-bin, then the breathing signal and heart signal are extracted by using the cascaded second-order filter. Finally, the filtered peak information of the heart signal is used to generate Gaussian pulse to simulate the ECG signal, and the LSTM network trained by the instantaneous frequency and spectral entropy features extracted from the 2017 PhysioNet/CinC challenge data is used to classify signals. The experimental results show that the detection algorithm can effectively detect the breath and heartbeat of human body, and recognize the AFib signal.

The dedicated memory of hardware accelerators can be insufficient to store all weights and/or intermediate states of large deep learning models. Although model parallelism is a viable approach to reduce the memory pressure issue, significant modification of the source code and considerations for algorithms are required. An alternative solution is to use out-of-core methods instead of, or in addition to, data parallelism. We propose a performance model based on the concurrency analysis of out-of-core training behavior, and derive a strategy that combines layer swapping and redundant recomputing. We achieve an average of 1.52x speedup in six different models over the state-of-the-art out-of-core methods. We also introduce the first method to solve the challenging problem of out-of-core multi-node training by carefully pipelining gradient exchanges and performing the parameter updates on the host. Our data parallel out-of-core solution can outperform complex hybrid model parallelism in training large models, e.g. Megatron-LM and Turning-NLG., ACM/IEEE Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC'20)

The primary objective of face morphing is to combine face images of different data subjects (e.g. a malicious actor and an accomplice) to generate a face image that can be equally verified for both contributing data subjects. In this paper, we propose a new framework for generating face morphs using a newer Generative Adversarial Network (GAN) - StyleGAN. In contrast to earlier works, we generate realistic morphs of both high-quality and high resolution of 1024$\times$1024 pixels. With the newly created morphing dataset of 2500 morphed face images, we pose a critical question in this work. \textit{(i) Can GAN generated morphs threaten Face Recognition Systems (FRS) equally as Landmark based morphs?} Seeking an answer, we benchmark the vulnerability of a Commercial-Off-The-Shelf FRS (COTS) and a deep learning-based FRS (ArcFace). This work also benchmarks the detection approaches for both GAN generated morphs against the landmark based morphs using established Morphing Attack Detection (MAD) schemes., Accepted in IWBF 2020

Federated learning is a recently proposed distributed machine learning paradigm for privacy preservation, which has found a wide range of applications where data privacy is of primary concern. Meanwhile, neural architecture search has become very popular in deep learning for automatically tuning the architecture and hyperparameters of deep neural networks. While both federated learning and neural architecture search are faced with many open challenges, searching for optimized neural architectures in the federated learning framework is particularly demanding. This survey paper starts with a brief introduction to federated learning, including both horizontal, vertical, and hybrid federated learning. Then neural architecture search approaches based on reinforcement learning, evolutionary algorithms and gradient-based are presented. This is followed by a description of federated neural architecture search that has recently been proposed, which is categorized into online and offline implementations, and single- and multi-objective search approaches. Finally, remaining open research questions are outlined and promising research topics are suggested.

The nonlinear relationship between influential factors and house price and inadequate number of sample size could be the cause of the poor performance of the traditional models. Meanwhile, the daily data of the real estate market is very huge and it is increasing rapidly. The traditional house price prediction approaches are lack of capacity for massive data analysis, causing low utilization of data. To address these concerns, a house price prediction model based on deep learning is proposed in this paper, implemented on the TensorFlow framework. Adam optimizer is used to train the model, where Relu function is adopted to be the activation function. Then house price trend is predicted based on the ARIMA model. By using Scrapy, housing data are obtained from Internet to be the experimental dataset. Comparative experiments were conducted between the proposed approach and SVR method. The experimental results show that individual house price predicted by the proposed approach is better than that of SVR method. And the predicted house price trend is mainly agreement with the real situation.

Gated recurrent unit (GRU) networks perform well in sequence learning tasks and overcome the problems of vanishing and explosion of gradients in traditional recurrent neural networks (RNNs) when learning long-term dependencies. Although they apply essentially to financial time series predictions, they are seldom used in the field. To fill this void, we propose GRU networks and its improved version for predicting trading signals for stock indexes of the Hang Seng Indexes (HSI), the Deutscher Aktienindex (DAX) and the S&P 500 Index from 1991 to 2017, and compare the GRU-based models with the traditional deep net and the benchmark classifier support vector machine (SVM). Experimental results show that the two GRU models proposed in this paper both obtain higher prediction accuracy on these data sets, and the improved version can effectively improve the learning ability of the model.

Motivation Genome-wide association studies are now shifting focus from analysis of common to rare variants. As power for association testing for individual rare variants may often be low, various aggregate level association tests have been proposed to detect genetic loci. Typically, power calculations for such tests require specification of large number of parameters, including effect sizes and allele frequencies of individual variants, making them difficult to use in practice. We propose to approximate power to a varying degree of accuracy using a smaller number of key parameters, including the total genetic variance explained by multiple variants within a locus. Results We perform extensive simulation studies to assess the accuracy of the proposed approximations in realistic settings. Using these simplified power calculations, we develop an analytic framework to obtain bounds on genetic architecture of an underlying trait given results from genome-wide association studies with rare variants. Finally, we provide insights into the required quality of annotation/functional information for identification of likely causal variants to make meaningful improvement in power. Availability and implementation A shiny application that allows a variety of Power Analysis of GEnetic AssociatioN Tests (PAGEANT), in R is made publicly available at https://andrewhaoyu.shinyapps.io/PAGEANT/. Supplementary information Supplementary data are available at Bioinformatics online.

More and more deep reinforcement learning algorithms have been proposed and demonstrated on a series of decision-making domains. With duality in deep reinforcement learning substantially summarized, this paper summarizes duality in deep reinforcement learning and proposes the duality with priority in Part I-Theory. It proposed an asynchronous architecture for dual priority in deep reinforcement learning. Experiments improve the state of the art on discrete action environments as well as continuous control tasks, while demonstrate its final performance and data efficiency.

In this paper, we propose a novel pretraining-based encoder-decoder framework, which can generate the output sequence based on the input sequence in a two-stage manner. For the encoder of our model, we encode the input sequence into context representations using BERT. For the decoder, there are two stages in our model, in the first stage, we use a Transformer-based decoder to generate a draft output sequence. In the second stage, we mask each word of the draft sequence and feed it to BERT, then by combining the input sequence and the draft representation generated by BERT, we use a Transformer-based decoder to predict the refined word for each masked position. To the best of our knowledge, our approach is the first method which applies the BERT into text generation tasks. As the first step in this direction, we evaluate our proposed method on the text summarization task. Experimental results show that our model achieves new state-of-the-art on both CNN/Daily Mail and New York Times datasets.

In this work, we focus on complex question semantic parsing and propose a novel Hierarchical Semantic Parsing (HSP) method, which utilizes the decompositionality of complex questions for semantic parsing. Our model is designed within a three-stage parsing architecture based on the idea of decomposition-integration. In the first stage, we propose a question decomposer which decomposes a complex question into a sequence of sub-questions. In the second stage, we design an information extractor to derive the type and predicate information of these questions. In the last stage, we integrate the generated information from previous stages and generate a logical form for the complex question. We conduct experiments on COMPLEXWEBQUESTIONS which is a large scale complex question semantic parsing dataset, results show that our model achieves significant improvement compared to state-of-the-art methods.

The success of portfolio construction depends primarily on the future performance of stock markets. Recent developments in machine learning have brought significant opportunities to incorporate prediction theory into portfolio selection. However, many studies show that a single prediction model is insufficient to achieve very accurate predictions and affluent returns. In this paper, a novel portfolio construction approach is developed using a hybrid model based on machine learning for stock prediction and mean–variance (MV) model for portfolio selection. Specifically, two stages are involved in this model: stock prediction and portfolio selection. In the first stage, a hybrid model combining eXtreme Gradient Boosting (XGBoost) with an improved firefly algorithm (IFA) is proposed to predict stock prices for the next period. The IFA is developed to optimize the hyperparameters of the XGBoost. In the second stage, stocks with higher potential returns are selected, and the MV model is employed for portfolio selection. Using the Shanghai Stock Exchange as the study sample, the obtained results demonstrate that the proposed method is superior to traditional ways (without stock prediction) and benchmarks in terms of returns and risks.

In a structural health monitoring (SHM) system that uses digital cameras to monitor cracks of structural surfaces, techniques for reliable and effective data compression are essential to ensure a stable and energy efficient crack images transmission in wireless devices, e.g., drones and robots with high definition cameras installed. Compressive sensing (CS) is a signal processing technique that allows accurate recovery of a signal from a sampling rate much smaller than the limitation of the Nyquist sampling theorem. The conventional CS method is based on the principle that, through a regularized optimization, the sparsity property of the original signals in some domain can be exploited to get the exact reconstruction with a high probability. However, the strong assumption of the signals being highly sparse in an invertible space is relatively hard for real crack images. In this paper, we present a new approach of CS that replaces the sparsity regularization with a generative model that is able to effectively capture a low dimension representation of targeted images. We develop a recovery framework for automatic crack segmentation of compressed crack images based on this new CS method and demonstrate the remarkable performance of the method taking advantage of the strong capability of generative models to capture the necessary features required in the crack segmentation task even the backgrounds of the generated images are not well reconstructed. The superior performance of our recovery framework is illustrated by comparing with three existing CS algorithms. Furthermore, we show that our framework is extensible to other common problems in automatic crack segmentation, such as defect recovery from motion blurring and occlusion., 34 pages, 15 figures, 3 tables

In modern high-performance computing (HPC) systems, users are usually requested to estimate the job runtime for system scheduling when they submit a job. In general, an underestimation of job runtime will cause the HPC system to terminate the job before its completion. If users could be notified that their jobs may not finish before its allocated time expires, users can take actions, such as killing the job and resubmitting it after parameter adjustment, to save time and cost. Meanwhile, the productivity of HPC systems could also be vastly improved. In this paper, we propose a data-driven approach – that is, one that actively observes, analyzes, and logs jobs – for predicting underestimation of job runtime on HPC systems. Using data produced by TSUBAME 2.5, a supercomputer deployed at the Tokyo Institute of Technology, we apply machine learning algorithms to recognize patterns about whether the underestimation of job runtime occurs. Our experimental results show that our approach on runtime-underestimation prediction with 80% precision, 70% recall and 74% F1-score on the entirety of a given dataset. Finally, we split the entire job data set into subsets categorized by scientific application name. The best precision, recall and F1-score of subsets on runtime-underestimation prediction achieved 90%, 95% and 92% respectively.

Training machine learning (ML) models with large datasets can incur significant resource contention on shared clusters. This training typically involves many iterations that continually improve the quality of the model. Yet in exploratory settings, better models can be obtained faster by directing resources to jobs with the most potential for improvement. We describe SLAQ, a cluster scheduling system for approximate ML training jobs that aims to maximize the overall job quality. When allocating cluster resources, SLAQ explores the quality-runtime trade-offs across multiple jobs to maximize system-wide quality improvement. To do so, SLAQ leverages the iterative nature of ML training algorithms, by collecting quality and resource usage information from concurrent jobs, and then generating highly-tailored quality-improvement predictions for future iterations. Experiments show that SLAQ achieves an average quality improvement of up to 73% and an average delay reduction of up to 44% on a large set of ML training jobs, compared to resource fairness schedulers., Appeared in the 1st SysML Conference. Full paper published in ACM SoCC 2017

Genome-wide association studies are now shifting focus from analysis of common to uncommon and rare variants with an anticipation to explain additional heritability of complex traits. As power for association testing for individual rare variants may often be low, various aggregate level association tests have been proposed to detect genetic loci that may contain clusters of susceptibility variants. Typically, power calculations for such tests require specification of large number of parameters, including effect sizes and allele frequencies of individual variants, making them difficult to use in practice. In this report, we approximate power to varying degree of accuracy using a smaller number of key parameters, including the total genetic variance explained by multiple variants within a locus. We perform extensive simulation studies to assess the accuracy of the proposed approximations in realistic settings. Using the simplified power calculation methods, we then develop an analytic framework to obtain bounds on genetic architecture of an underlying trait given results from a genome-wide study and observe important implications for the completely lack of or limited number of findings in many currently reported studies. Finally, we provide insights into the required quality of annotation/functional information for identification of likely causal variants to make meaningful improvement in power of subsequent association tests. A shiny application, Power Analysis for GEnetic AssociatioN Tests (PAGEANT), in R implementing the methods is made publicly available.

The learning problem of ranking arises in many tasks, including the question answering, information retrieval, and movie recommendation. In these tasks, the ordering of the answers, documents or movies returned is a critical aspect of the system. Recently, deep learning approaches have gained a lot of attention from the research community and industry for their ability to automatically learn optimal feature representation for a given task. We aim to solve the answer ranking problem in practical question answering system with deep learning approaches. In this paper, we define a composite representation for questions and answers by combining convolutional neural network (CNN) with bidirectional long short-term memory (biLSTM) models, and learn a similarity function to relate them in a supervised way from the available training data. Considering the limited training data, we propose a hypernym strategy to get more general text pairs and test the effectiveness of different strategies. Experimental results on a public benchmark dataset from TREC demonstrate that our system outperforms previous work which requires syntactic features and some deep learning models.

In microwave imaging for brain cancer detection, signals are generally degraded by noise. In this paper, we investigate the use of Discrete Wavelet Transform (DWT) based signal processing to improve the noise performance of an UWB based microwave imaging system for brain cancer detection. To test the noise suppression properties of the DWT, firstly, Gaussian white noise is added to the received pulse in a simulated microwave imaging system, such that the signal-to-noise ratios (SNRs) are 60dB and 45dB, respectively. These noisy signals are then processed and de-noised using the DWT. The de-noised signals are used to create cross-sectional images of a cancerous brain model, with the tumour highlighted. These resulting images demonstrate the validity of a DWT based de-noising method for brain cancer detection.

This paper proposes a novel method for reconstruction of the target three-dimensional (3D) objects based on motion vision. The main contribution of this paper is twofold: (1) we build a new mathematical model to express the position relationship of the camera, the image plane, the ground plane and the target 3D objects at different times. (2) We propose and implement a 3D reconstruction algorithm refer to the mathematical model using geometric theory. With the result of that, we know the distance from the camera to the objects and the height of the objects in the real world. The experimental indicates have shown the accuracy of the approach presented in this paper.

Face morphing attacks target to circumvent Face Recognition Systems (FRS) by employing face images derived from multiple data subjects (e.g., accomplices and malicious actors). Morphed images can be verified against contributing data subjects with a reasonable success rate, given they have a high degree of facial resemblance. The success of morphing attacks is directly dependent on the quality of the generated morph images. We present a new approach for generating strong attacks extending our earlier framework for generating face morphs. We present a new approach using an Identity Prior Driven Generative Adversarial Network, which we refer to as MIPGAN (Morphing through Identity Prior driven GAN) . The proposed MIPGAN is derived from the StyleGAN with a newly formulated loss function exploiting perceptual quality and identity factor to generate a high quality morphed facial image with minimal artefacts and with high resolution. We demonstrate the proposed approach’s applicability to generate strong morphing attacks by evaluating its vulnerability against both commercial and deep learning based Face Recognition System (FRS) and demonstrate the success rate of attacks. Extensive experiments are carried out to assess the FRS’s vulnerability against the proposed morphed face generation technique on three types of data such as digital images, re-digitized (printed and scanned) images, and compressed images after re-digitization from newly generated MIPGAN Face Morph Dataset . The obtained results demonstrate that the proposed approach of morph generation poses a high threat to FRS.