Search

Twitter spam classification is a tough challenge for social media platforms and cyber security companies. Twitter spam with illegal links may evolve over time in order to deceive filtering models, causing disastrous loss to both users and the whole network. We define this distributional evolution as a concept drift scenario. To build an effective model, we adopt K-L divergence to represent spam distribution and use a multiscale drift detection test (MDDT) to localize possible drifts therein. A base classifier is then retrained based on the detection result to gain performance improvement. Comprehensive experiments show that K-L divergence has highly consistent change patterns between features when a drift occurs. Also, the MDDT is proved to be effective in improving final classification result in both accuracy, recall, and f-measure.

As it is difficult to directly share data among different organizations, data federation brings new opportunities to the data-related cooperation among different organizations by providing abstract data interfaces. With the development of Cloud computing, organizations store data on the Cloud to achieve elasticity and scalability for data processing. The existing data placement approaches generally only consider one aspect, which is either communication cost or time cost, and do not consider the features of jobs that process the data. In this paper, we propose an approach to enable secure data processing on the Cloud with the data from different organizations. The approach consists of a data federation platform for secure data processing on the Cloud named FedCube and a greedy data placement algorithm that creates a plan to store data on the Cloud in order to achieve multiple objectives based on a cost model. The cost model is composed of two objectives, i.e., reducing both monetary cost and execution time. We present an experimental evaluation by comparing our data placement algorithm with the existing methods based on the data federation platform. The experiments show that our proposed algorithm significantly reduce the total cost (up to 69.8%).

A crane system often works in a complex environment. Traditional system identification methods are difficult to accurately model crane system identification's real dynamic performance, whether online or offline. Therefore, a Data-driven model predictive control (D-MPC) algorithm is proposed in this paper. Based on Bayesian optimization, the crane's local linear dynamic model is learned to improve the rapidly anti-swing and precise positioning. By collecting data through closed-loop experiments and using Bayes to construct the Gaussian model for guiding learning, the controller parameters and prediction model with the best closed-loop performance can be found. Simulation results show that we explicitly find the dynamics model that produces the best control performance for the actual system, and the method can quickly suppress the swing and realize the accurate trolley positioning. The results verify the proposed control algorithm's effectiveness, feasibility and superior performance of the proposed method by comparing it with double-closed-loop proportional–integral– derivative (PID).

Imbalanced classification using a support vector machine (SVM) is a normal but crucial problem in machine learning. Compared with binary classification, multiclass classification is much more complicated. Most existing studies on imbalanced classification using SVM focus on binary imbalanced classification; while only few of them look into imbalanced classification with multiple classes. Pre-clustering is a useful technique to prepare proper data from an imbalanced dataset for a classifier. It can be used to extract the feature of a dataset first and improve classification performance. Density peak based on Euclidean distance proves its effectiveness and generality in clustering. Motivated by this and the fact that the number of clusters is known in multi-class classification using a one-vs-rest strategy, we combine density peak clustering and SVM to propose a new pre-clustering method to perform effective imbalanced classification with multiple classes. Specifically, we transform a multi-class classification problem into several binary classification tasks. The results on 5 public datasets in terms of F-measure, G-mean and Area Under Curve (AUC) show its superiority over the original SVM and SVM with other methods including random under-sampling, Synthetic Minority Oversampling Technique, pre-clustering using K-Means and EasyEnsemble methods using either a one-vs-rest or one-vs-one strategy.

Question Generation aims to create various questions from a given passage, which can provide education material, improve the training of question answering, and help chat bots have cold-to-start or continue to talk to people. Chinese Question Generation is a new research area, which mainly uses a rule-based model to convert declarative sentences into questions. We propose a Stroke-Aware Copy Network (SAC-Net), which is a sequence-to-sequence model. It can enhance the performance of Chinese question generation by using various specific components and techniques for Chinese characteristics. In view of the fact that the morphology of Chinese characters is related to its meaning, this research introduces the information of strokes in Chinese and obtain the relationship among strokes in Chinese characters. This model introduces the direction information of the questions, which helps us generate more diverse questions. For the out-of-vocabulary (OOV) word problem that often occurs in Chinese, this research has improved a Self-attention Copy mechanism. Extensive evaluations confirm that this model improves the performance of Chinese problem generation significantly and represents a state-of-the-art neural Chinese problem generation model.

Due to the absence of equipment operation parameters or inaccurate input of basic information, it is impossible to accurately grasp the operation status of transformers, resulting in the current situation of group transformers after fault repair became serious rapidly. This paper proposes a general method of grading management, which is easy for operators to accurately diagnose the operation status of the transformers. By supplementing the basic data such as bad operating conditions and family defects of transformers, screened out the transformers that have experienced more than 70% short-circuit current impulse, and the changing trend of historical test data of both capacitance and reactance is selected as the preferred condition for judging the winding deformation position, and constructed the capacitance-reactance models to equivalent the deformation amount. Through the analysis of historical data of group transformers, the latent danger of non-blackout equipment can be found in advance, and the passive situation of transformer maintenance after long-term faults can be changed.

Nowadays plenty of pointer meters are used in the field of chemical industry and electrical power system. To avoid reading their indication manually, many algorithms based on computer vision have been proposed to read pointer meters automatically. These methods, however, are limited to meters whose scales are uniform, and their accuracy is vulnerable to the error in the recognition of a meter’s center. In this paper, a novel automatic reading algorithm of pointer meters based on scale seeking is proposed to overcome the weaknesses of the existing methods. Differing from the popular angle-based methods, we obtain the indication of the meter by comparing the distances between the peak of pointer and its nearest scales. The position and values of all scales can be automatically acquired and inferred by using our scale seeking and value inference algorithms, which is independent of any prior information in a database. Experiments prove that the algorithm can be applied to both meters with uniform or non-uniform scales effectively.

Word representations are crucial for many nature language processing tasks. Most of the existing approaches learn contextual information by assigning a distinct vector to each word and pay less attention to morphology. It is a problem for them to deal with large vocabularies and rare words. In this paper we propose an Adaptive Wordpiece Language Model for learning Chinese word embeddings (AWLM), as inspired by previous observation that subword units are important for improving the learning of Chinese word representation. Specifically, a novel approach called BPE+ is established to adaptively generates variable length of grams which breaks the limitation of stroke n-grams. The semantical information extraction is completed by three elaborated parts i.e., extraction of morphological information, reinforcement of fine-grained information and extraction of semantical information. Empirical results on word similarity, word analogy, text classification and question answering verify that our method significantly outperforms several state-of-the-art methods.

Image demosaicking is the most crucial preprocessing step in the current color digital camera pipeline. Efficiency and high quality are of importance to demosaicking methods at the request of practical applications. Recently, convolutional neural network (CNN) has demonstrated its superior performance on image demosaicking. However, most existed CNN-based demosaicking methods fail to take full advantage of the self-similarity and redundancy in natural image, and interpolation artifacts (e.g. zippering and color moire) easily occur when local geometry cannot be inferred correctly from neighboring pixels. To solve these problems, we propose a fully convolutional feature pyramid network to exploit image self-similarity and redundant information as much as possible for image demosaicking. Furthermore, we add a compact channel attention module to the proposed network to flexibly rescale channel-wise features by modeling interdependencies among channels. Our experimental results on three datasets show that our method obviously outperforms state-of-the-art methods on both quantitative and visual quality assessments, and maintains competitive running time in the inference stage.

One of the difficulties in computer vision is how to build an accurate classifier for a new target domain with insufficient labeled images from a related source domain with labeled images. Adversarial learning is a novel domain adaptation method that tackles this challenge by training robust deep networks and reducing the distribution difference between source and target domains, thus improving the classification performance on a target task. However, most prior adversarial adaptation learning approaches merely reduce the distribution difference across domains through GAN (Generative Adversarial Networks)-based loss, but when the performance of a generator or discriminator in GAN is degraded, the distribution difference between source and target domains are difficult to decrease. In this paper, we propose a novel generalized framework for adversarial domain adaptation, referred to as Generative Adversarial Distribution Matching. Our idea is to add the data discrepancy distance between source and target domains to the objective function of the generator so as to reduce distribution difference across domains through a Generator and a Discriminator compete against each other. Comprehensive experimental results confirm that it can well outperform several state-of-the-art methods for cross-domain image classification problems.

Photovoltaic (PV) power generation utilizing clean solar energy is increasingly conducive to relieving energy crisis and environmental pollution. Affected by the fluctuation and uncertainty of meteorological factors, short-term PV power is volatile in nature, posing threats to power supply reliability and stability. Consequently, accurate PV production forecasting plays a vital role in steadily running and managing a power system. However, due to the intrinsic characteristics of variability and fluctuation in PV data, it is challenging to get acceptable output prediction via conventional regression methods. Moreover, the raw data in our regression task originates from a PV plant whose stored PV values are hierarchically non-continuous with conspicuously diverse classes, making it even harder to conduct precise prediction. In this paper, we propose a tree-based prediction model based on the XGBoost regression algorithm. The experimental results show that the proposed prediction model achieves the highest average forecasting accuracy and stable generalization performance, indicating its validity for hierarchically non-continuous short-term PV output prediction.

Imbalance classification is a common but critical problem in machine learning and artificial intelligence. Derived from structural risk minimization, a Support Vector Machine (SVM) enjoys great reputation in classification. However, the original SVM is not suitable for the imbalance classification and the existing modifications of SVM for this kind of problems fail to take the distribution of datasets into full consideration, thereby leading to some remarkable loss in their classification performance. Recently, an Adaptive Clustering by Fast Search and Find of Density Peaks (ADPclust) is proposed and performs well in finding cluster centroids in a sample space automatically and more reliably by using adaptive density peak detection and silhouette theory. Motivated by this, this work proposes an adaptive pre-clustering SVM (AP-SVM) such that the information of the original dataset distribution is well utilized to yield balanced sub-datasets for accurate and efficient classification. Specifically, AP-SVM clusters the majority into several groups given a dataset and then applies undersampling on every cluster to re-balance the dataset to be used in the SVM classification step. After experiments on 10 binary public datasets and evaluation using Area Under Curve (AUC), F-Measure, G-Mean, we well show the superiority of the proposed method over SVM, Synthetic Minority Over-sampling Technique algorithm (SMOTE), Undersampling-SVM (U-SVM), K-Means, Fuzzy C Means and EasyEnsemble.

Concept drifts can occur in data streams, which disable any models built on static data distribution to fit dynamic or recurrent concepts. If there is a feature that can monitor the stableness of such distribution, then we have a proper reference to adapt the model. Based on this idea, we propose a novel approach named Multiscale Drift Detection Test (MDDT) that localizes abrupt drift points when detection feature values fluctuate. MDDT is based on a resampling scheme and a paired student t-test. It applies a detection procedure on a broad and a narrow scale. This multiscale structure not only reduces massive time of a constant checking process, but also filters noise in the detection features. Experiments are performed via synthetic and real-world datasets. The results indicate that the proposed method outperforms the state-of-art algorithms in terms of computation cost and average accuracy.

This paper described an experiment which was designed to validate the effectiveness of an aerodynamic olfactory effect model for the simulation of gas-sensitive rotorcrafts. The model simulates the rotor wakes using free vortices and calculates the distortion of the gas plume using filament based plume model. We designed an experiment in which a camera recorded the distortion of the smoke plume, and proposed a practical video processing routine to extract the smoke dispersion trajectory. Then the trajectory was compared with the simulated one to validate the model. Results showed that this aerodynamic olfactory effect model is a credible way to simulate odor diffusion process under the influence of quadrotor.

Accurate pose (position and attitude) measurement is a key for automatic docking of cabins. In this paper, a stereo vision based cabin's six-dimensional (6D: three translational positions plus three rotational angles) pose measurement method has been proposed. First, the feature points provided by the visual marker pasted on each cabin's outer surfaces are extracted. The three-dimensional translational coordinates of the feature points in the camera coordinate system are obtained by the stereo vision system. Then the coordinate systems of two markers on two cabins are established through the feature points. The two cabins' 6D poses are denoted by two pose matrices calculated from the two markers' coordinate systems. Relative pose between two cabins is obtained through the two pose matrices. The pose measurement method has been verified using a six-degree-of-freedom (DOF) motion platform. Experimental results show that the errors in three translations and three rotations are ±0.06 mm out of ±5 mm and ±0.07° out of ±3°, respectively.

The purpose of this paper is to develop and validate a configurable system that can wirelessly acquire gastric electrical activity called slow waves, and deliver high energy electrical pulses to modulate its activity. The system is composed of a front-end unit, and an external stationary backend unit that is connected to a computer. The front-end unit contains a recording module with four channels, and a stimulating module with two channels. Commercial off-theshelf components were used to develop front- and back-end units. A graphical user interface (GUI) was designed in LabVIEW to process and display the recorded data in realtime, and store the data for off-line analysis. Besides, the gain of the analog conditioning circuit as well as the stimulation pulse configuration is programmable directly through the GUI. The system was successfully validated on bench top. The benchtop studies showed an appropriate frequency response for analog conditioning and digitization resolution to acquire gastric slow waves. Moreover, the system was able to deliver electrical pulses at amplitudes up to ±24 mA and ±12 mA to a load of up to 0.5 k $\Omega $ and 1 $\textbf{k}\Omega $, respectively. This study reports the first high-energy stimulator that can be controlled wirelessly and integrated into a gastric bioelectrical activity monitoring system. The system can be used for treating functional gastrointestinal disorders.

One of the challenges in computer vision is how to learn an accurate classifier for a new domain by using labeled images from an old domain under the condition that there is no available labeled images in the new domain. Domain adaptation is an outstanding solution that tackles this challenge by employing available source-labeled datasets, even with significant difference in distribution and properties. However, most prior methods only reduce the difference in subspace marginal or conditional distributions across domains while completely ignoring the source data label dependence information in a subspace. In this paper, we put forward a novel domain adaptation approach, referred to as Enhanced Subspace Distribution Matching. Specifically, it aims to jointly match the marginal and conditional distributions in a kernel principal dimensionality reduction procedure while maximizing the source label dependence in a subspace, thus raising the subspace distribution matching degree. Extensive experiments verify that it can significantly outperform several state-of-the-art methods for cross-domain image classification problems.

Particle Swarm Optimization (PSO) is a population-based algorithm designed to tackle various optimization problems. However, its performance deteriorates significantly when optimization problems are subjected to noise. PSO is strongly influenced by each particle's own previous best one and global best one, which may lead to premature convergence and fall into local optima. This also holds true for various PSO variants dealing with optimization problems in noisy environments. Opposition-based learning (OBL) is well-known for its ability to increase population diversity. In this paper, we propose hybrid PSO algorithms that introduce OBL into a PSO variant for improving the latter's performance. The proposed hybrid algorithm employs probabilistic OBL for particle swarm. In contrast to other integrations of PSO and OBL, we select the top fittest particles from the current swarm and its opposite swarm to improve the entire swarm's fitness. Experiments on 10 benchmark functions subject to different levels of noise show that the proposed algorithm has better performance in most cases.

Many real-world datasets suffer from the problem of class imbalance, i.e., they have a minority class being only a small portion of the whole dataset. Under-sampling techniques, e.g., EasyEnsemble (EE), present an efficient approach to imbalanced classification problems. However, imbalance is not the only factor that harms the performance of conventional classifiers. The presence of noises is another issue that really complicates the process of classifier learning. This paper presents a new noise-filtered under-sampling algorithm by incorporating an Iterative-Partitioning Filter (IPF) into EE, named EE-IPF for short. IPF can remove noises from both majority and minority classes. Comprehensive experiments are performed to test its performance via eleven commonly-used benchmark datasets. The results show its outstanding performance in terms of popular ly-used metrics for imbalanced classification.

MSR Image Recognition Challenge (IRC) 2016 tries to deal with designing a dog breeds recognition system based on Clickture-Dog dataset. To address the task, we proposed an effective method with systematic strategies as follows. We presented a data cleaning method using faster-rcnn to learn a dog detector. Besides, we ensembeled a series of CNN models to enhance the robustness. A dense evaluation strategy was carried out to boost the test accuracy. Finally, we achieved the first prize of this challenge with 89.65% top-5 accuracy.

As the supply potential has declined gradually and the pressure for better environment intensifies, the demand for clean energy arises continuously. The consumptive demand for Plug-in Hybrid Electric Vehicle (PHEV) thus increases. However, load peak caused by their disordered charging can be detrimental to an entire power grid. Several methods have been proposed to establish ordered PHEV charging. While focusing on single-objective optimal load scheduling, they fail to meet the real requirements for efficient multiple objective optimization. This work proposes a novel weight-aggregation based multi-objective particle swarm optimization method to solve the load scheduling problem. Its effectiveness and efficiency to generate a Pareto front of this problem are verified and compared with those of the state-of-the-art approaches.

The track 2 and track 3 of ChaLearn 2016 can be considered as Multi-Label Classification problems. We present a framework of learning deep binary encoding (DeepBE) to deal with multi-label problems by transforming multi-labels to single labels. The transformation of DeepBE is in a hidden pattern, which can be well addressed by deep convolutions neural networks (CNNs). Furthermore, we adopt an ensemble strategy to enhance the learning robustness. This strategy is inspired by its effectiveness in fine-grained image recognition (FGIR) problem, while most of face related tasks such as track 2 and track 3 are also FGIR problems. By DeepBE, we got 5.45% and 10.84% mean square error for track 2 and track 3 respectively. Additionally, we proposed an algorithm adaption method to treat the multiple labels of track 2 directly and got 6.84% mean square error.

Bat algorithm is a novel swarm intelligent algorithm inspired by the echolocation behavior of bats. However, the exploitation capability is not well for some multi-model problems. In this paper, a new parameter, inertia weight is introduced to improve the local search capability. In this new variant, the inertia weight is used to control the influence of the inertia of previous velocity for each bat. To test the performance, the CEC2013 shifted and rotated benchmark problems are used to compare, simulation results show our method is validity.

In this paper, the reliability evaluation of distribution network is studied by using the sequential Monte Carlo simulation based on dual sampling. A reliability evaluation model of distribution network is established. The range influenced by fault is determined by the analysis method of node fault. And then, all kinds of reliability index are calculated. With Matlab programming, IEEE RBTS Bus 6 test system is evaluated to prove that the method in this paper is effective. Then the simulation analysis is carried out for the cases that whether there is a standby transformer in power distribution network, whether the circuit breaker is reliable and whether there is a contact switch between the two feeders. The results show that the reliability of distribution network system can be enhanced by adding the standby transformer and the contact switch and improving the reliability of the circuit breaker.

Gonzalez Vasco et al. extend the area of application of algorithms for the hidden number problem in 2004. Using this extension and relations among the bits in and binary fraction expansion of x mod p/p, we present a probabilistic algorithm for some trapdoor functions to recover a hidden message when an imperfect oracle is given of predicting most significant bits in hidden message. We show that computing the most significant bit in message encrypted by ElGmal encryption function is as hard as computing the entire plaintext, and so is RSA.

Clustering analysis is an important tool of data mining. The study on efficient clustering has great significance, especially in improving a clustering algorithm's adaptability and usefulness. Clustering ensemble (CE) integrates several clustering algorithms such that the clustering results can be effectively improved. This work investigates similarity-based methods and proposes a new method called weight- incorporated similarity-based clustering ensemble (WSCE). Six classic data sets are used to test single clustering algorithms, similarity-based one, and the proposed one via simulation. The results prove the validity and performance advantage of the proposed method.