Improving Classification Efficiency of Fake News using Semi-Supervised Method

Online News media which is more accessible, cheaper, and faster to consume, is also of questionable quality as there is less moderation. Anybody with a computing device and internet connection can take part in creating, contributing, and spreading news in online portals. Social media has intensified the problem further. Due to the high volume, velocity, and veracity, online news content is beyond traditional moderation, also known as moderation through human experts. So different machine learning method is being tested and used to spot fake news. One of the main challenges for fake-news classification is getting labeled instances for this high volume of real-time data. In this study, we examined how semi-supervised machine learning can help to decrease the need for labeled instances with an acceptable drop of accuracy. The accuracy difference between the supervised classifier and the semi-supervised classifier is around 0.05 while using only five percent of label instances of the supervised classifier. We tested with logistic regression, SVM, and random forest classifier to prove our hypothesis.

Analysis of Influence of Different Parameters on Numerical Simulation of NACA0012 Incompressible External Flow Field under High Reynolds Numbers

Currently, influence analysis of simulation parameters, especially the trailing edge shape and the corresponding modeling method on the force coefficients of NACA0012 under a high Reynolds number, is relatively sparse. In this paper, two trailing edge shapes are designed by three modeling methods and combined with three far-field distances to establish eighteen two-dimensional external flow fields. The same number of structured grids are generated by a unified grid strategy and the SST k-omega and the Spalart–Allmaras models are adopted to solve the NS equations to realize the numerical simulations. Unlike under low Reynolds numbers, the analysis results show that although the accuracy difference between the sharp trailing edge and the blunt trailing edge decreases as the attack angle range increases, the former is preferred in all studied ranges. As to the corresponding modeling methods, the NACA4 and the definition formula are preferred, the choice of which depends on the studied range. In particular, a greater number of data points adopted into the definition formula is not necessarily better. Considering the error ratios comprehensively, the simulation configurations of sharp trailing edge + 20 m far-field distance + SA/SST/SST/SST/SST/SA turbulence model obtains optimal simulation effects.

Learning Realistic Patterns from Visually Unrealistic Stimuli: Generalization and Data Anonymization

Good training data is a prerequisite to develop useful Machine Learning applications. However, in many domains existing data sets cannot be shared due to privacy regulations (e.g., from medical studies). This work investigates a simple yet unconventional approach for anonymized data synthesis to enable third parties to benefit from such anonymized data. We explore the feasibility of learning implicitly from visually unrealistic, task-relevant stimuli, which are synthesized by exciting the neurons of a trained deep neural network. As such, neuronal excitation can be used to generate synthetic stimuli. The stimuli data is used to train new classification models. Furthermore, we extend this framework to inhibit representations that are associated with specific individuals. We use sleep monitoring data from both an open and a large closed clinical study, and Electroencephalogram sleep stage classification data, to evaluate whether (1) end-users can create and successfully use customized classification models, and (2) the identity of participants in the study is protected. Extensive comparative empirical investigation shows that different algorithms trained on the stimuli are able to generalize successfully on the same task as the original model. Architectural and algorithmic similarity between new and original models play an important role in performance. For similar architectures, the performance is close to that of using the original data (e.g., Accuracy difference of 0.56%-3.82%, Kappa coefficient difference of 0.02-0.08). Further experiments show that the stimuli can provide state-ofthe-art resilience against adversarial association and membership inference attacks.

Development and evaluation of the refined zenith tropospheric delay (ZTD) models

The tropospheric delay is a significant error source in Global Navigation Satellite System (GNSS) positioning and navigation. It is usually projected into zenith direction by using a mapping function. It is particularly important to establish a model that can provide stable and accurate Zenith Tropospheric Delay (ZTD). Because of the regional accuracy difference and poor stability of the traditional ZTD models, this paper proposed two methods to refine the Hopfield and Saastamoinen ZTD models. One is by adding annual and semi-annual periodic terms and the other is based on Back-Propagation Artificial Neutral Network (BP-ANN). Using 5-year data from 2011 to 2015 collected at 67 GNSS reference stations in China and its surrounding regions, the four refined models were constructed. The tropospheric products at these GNSS stations were derived from the site-wise Vienna Mapping Function 1 (VMP1). The spatial analysis, temporal analysis, and residual distribution analysis for all the six models were conducted using the data from 2016 to 2017. The results show that the refined models can effectively improve the accuracy compared with the traditional models. For the Hopfield model, the improvement for the Root Mean Square Error (RMSE) and bias reached 24.5/49.7 and 34.0/52.8 mm, respectively. These values became 8.8/26.7 and 14.7/28.8 mm when the Saastamoinen model was refined using the two methods. This exploration is conducive to GNSS navigation and positioning and GNSS meteorology by providing more accurate tropospheric prior information.

Gait Analysis Accuracy Difference with Different Dimensions of Flexible Capacitance Sensors

Stroke causes neurological pathologies, including gait pathologies, which are diagnosed by gait analysis. However, existing gait analysis devices are difficult to use in situ or are disrupted by external conditions. To overcome these drawbacks, a flexible capacitance sensor was developed in this study. To date, a performance comparison of flexible sensors with different dimensions has not been carried out. The aim of this study was to provide optimized sensor dimension information for gait analysis. To accomplish this, sensors with seven different dimensions were fabricated. The dimensions of the sensors were based on the average body size and movement range of 20- to 59-year-old adults. The sensors were characterized by 100 oscillations. The minimum hysteresis error was 8%. After that, four subjects were equipped with the sensor and walked on a treadmill at a speed of 3.6 km/h. All walking processes were filmed at 50 fps and analyzed in Kinovea. The RMS error was calculated using the same frame rate of the video and the sampling rate of the signal from the sensor. The smallest RMS error between the sensor data and the ankle angle was 3.13° using the 49 × 8 mm sensor. In this study, we confirm the dimensions of the sensor with the highest gait analysis accuracy; therefore, the results can be used to make decisions regarding sensor dimensions.

On fourth order accuracy stable difference scheme for a multi-point overdetermined elliptic problem

In this paper fourth order of accuracy difference scheme for approximate solution of a multi-point elliptic overdetermined problem in a Hilbert space is proposed. The existence and uniqueness of the solution of the difference scheme are obtained by using the functional operator approach. Stability, almost coercive stability, and coercive stability estimates for the solution of difference scheme are established. These theoretical results can be applied to construct a stable highly accurate difference scheme for approximate solution of multi-point overdetermined boundary value problem for multidimensional elliptic partial differential equations.

DEEPLIO: DEEP LIDAR INERTIAL SENSOR FUSION FOR ODOMETRY ESTIMATION

Having a good estimate of the position and orientation of a mobile agent is essential for many application domains such as robotics, autonomous driving, and virtual and augmented reality. In particular, when using LiDAR and IMU sensors as the inputs, most existing methods still use classical filter-based fusion methods to achieve this task. In this work, we propose DeepLIO, a modular, end-to-end learning-based fusion framework for odometry estimation using LiDAR and IMU sensors. For this task, our network learns an appropriate fusion function by considering different modalities of its input latent feature vectors. We also formulate a loss function, where we combine both global and local pose information over an input sequence to improve the accuracy of the network predictions. Furthermore, we design three sub-networks with different modules and architectures derived from DeepLIO to analyze the effect of each sensory input on the task of odometry estimation. Experiments on the benchmark dataset demonstrate that DeepLIO outperforms existing learning-based and model-based methods regarding orientation estimation and shows a marginal position accuracy difference.

iRG-4mC: Neural Network Based Tool for Identification of DNA 4mC Sites in Rosaceae Genome

DNA N4-Methylcytosine is a genetic modification process which has an essential role in changing different biological processes such as DNA conformation, DNA replication, DNA stability, cell development and structural alteration in DNA. Due to its negative effects, it is important to identify the modified 4mC sites. Further, methylcytosine may develop anywhere at cytosine residue, however, clonal gene expression patterns are most likely transmitted just for cytosine residues in strand-symmetrical sequences. For this reason many different experiments are introduced but they proved not to be viable choice due to time limitation and high expenses. Therefore, to date there is still need for an efficient computational method to deal with 4mC sites identification. Keeping it in mind, in this research we have proposed an efficient model for Fragaria vesca (F. vesca) and Rosa chinensis (R. chinensis) genome. The proposed iRG-4mC tool is developed based on neural network architecture with two encoding schemes to identify the 4mC sites. The iRG-4mC predictor outperformed the existing state-of-the-art computational model by an accuracy difference of 9.95% on F. vesca (training dataset), 8.7% on R. chinesis (training dataset), 6.2% on F. vesca (independent dataset) and 10.6% on R. chinesis (independent dataset). We have also established a webserver which is freely accessible for the research community.

Numerical dissipation of flux splitting schemes for contact discontinuities

The contact discontinuity is simulated by three kinds of flux splitting schemes to evaluate and analyse the influence of numerical dissipation in this paper. The numerical results of one-dimensional contact discontinuity problem show that if the flow velocity on both sides of the contact discontinuity is not simultaneously supersonic, the non-physical pressure and velocity waves may occur when the initial theoretically contact discontinuity is smeared into a transition zone spanning several grid-cells caused by numerical dissipations. Since these non-physical waves have no effect on the corresponding density dissipation, this paper considers these fluctuations as only numerical errors and are not part of the numerical dissipation. In addition, for two-dimensional flow field, the characteristics of high-order accuracy difference schemes, i.e. low dissipation and high resolution, may induce the multi-dimensional non-physical waves that interfere with each other to produce more complex non-physical flow structures, so the fluctuations in the calculated results should be treated with caution.

Compensation method for complex hysteresis characteristics on piezoelectric actuator based on Separated Level-loop Prandtl–Ishlinskii model

Piezoelectric ceramic actuators exhibit nonlinear hysteresis characteristics owing to their material properties. To modify the inverse piezoelectric effect as an ideal linear execution, the classical Prandtl–Ishlinskii (PI) model is usually used for compensation by feedforward control. The PI model compensates well on simple hysteresis characteristics. However, when the output requirements are complex, the PI model demonstrates uneven compensation accuracy on the complex hysteresis characteristics and cannot achieve an accuracy similar to that of simple hysteresis. This paper proposes a simplification of complex hysteresis: Separated Level-loop PI (SLPI) model. First, we use a loop-separation logic algorithm to simplify the complex hysteresis characteristics to obtain hysteresis in the form of single loops with loop levels and vertexes. Second, the hysteresis characteristics of each loop are independently modeled using the PI model. Finally, the inverse model is reconstructed using a rollback method to restore a positive sequence of the feedforward voltage; then, the feedforward voltage is input as a compensation value to achieve higher and more uniform accuracy. Experiments and discussions show that the SLPI model can effectively improve the compensation results of complex hysteresis characteristics; moreover, the average compensation accuracy difference between single hysteresis loops was reduced.