Search

Multilingual BERT (mBERT), a language model pre-trained on large multilingual corpora, has impressive zero-shot cross-lingual transfer capabilities and performs surprisingly well on zero-shot POS tagging and Named Entity Recognition (NER), as well as on cross-lingual model transfer. At present, the mainstream methods to solve the cross-lingual downstream tasks are always using the last transformer layer's output of mBERT as the representation of linguistic information. In this work, we explore the complementary property of lower layers to the last transformer layer of mBERT. A feature aggregation module based on an attention mechanism is proposed to fuse the information contained in different layers of mBERT. The experiments are conducted on four zero-shot cross-lingual transfer datasets, and the proposed method obtains performance improvements on key multilingual benchmark tasks XNLI (+1.5 %), PAWS-X (+2.4 %), NER (+1.2 F1), and POS (+1.5 F1). Through the analysis of the experimental results, we prove that the layers before the last layer of mBERT can provide extra useful information for cross-lingual downstream tasks and explore the interpretability of mBERT empirically., Accepted by ICPR 2022

In this paper, attitude stabilization control for flexible spacecraft subject to external disturbance without angular velocity and flexible mode variables measurement is considered. First, an adaptive law is constructed to estimate the upper bound of a lumped disturbance consisting of flexible accessories and external disturbance. Based on the proposed adaptive law, an adaptive observer is established to estimate angular velocity and the lumped disturbance. Besides, for the case where flexible modal variables cannot be measured, a flexible modal variable observer-based sliding mode control law is proposed. Simulations are performed to verify the validity of the proposed control law. The simulation results show that the proposed control law can eliminate the influence of external disturbance for flexible spacecraft under the situation that angular velocity and flexible modal variables cannot be measured.

In this paper, the attitude tracking control problem for rigid spacecraft is investigated in the presence of parameter uncertainties and external disturbances. An extended state observer is designed to estimate the parameter uncertainties and external disturbances. Then, by using the obtained estimation, a robust finite-time controller is proposed to achieve attitude tracking for the rigid spacecraft via the backstepping technique. The practical finite-time stability of the closed-loop system under the presented controller is proven by the Lyapunov stability theory. Finally, some simulation results are provided to illustrate the effectiveness and superiority of the proposed controller compared with some existing control schemes.

In this paper, the concept of fundamental matrices is generalized to discrete-time systems with state delays, and a simple property is derived. With the help of fundamental matrices, a prediction scheme is proposed for discrete-time systems with delays in both state and input.

In this paper, a terminal sliding mode controller based on disturbance observer is designed for rigid spacecraft. A disturbance observer is first proposed for the unmeasurable disturbance. The disturbance estimation error converges to zero in finite time. By combining this proposed observer with the sliding mode control technique, then a finite time attitude control law is presented for the rigid spacecraft. The variable boundary layer saturation function is used in the design of the switching portion of the terminal sliding mode controller to reduce chattering. The convergence of the observer and closed-loop system is proved by the Lyapunov stability theory.

This technical paper provides an attitude tracking strategy by employing a nonlinear model predictive controller (NMPC). To directly use the relative navigation information, the piecewise affine (PWA) model based on the Euler dynamics is adopted. Due to the singularity of Euler angles, a singularity-free strategy is proposed to complete continuous attitude tracking. Consider the sensor uncertainty and disturbance, the Kalman filter is combined with the NMPC framework. Finally, the numerical results are presented to show the control performance.

In this paper, the master equations for a two-level system driven by three photons has been derived. Particularly, the incident photons are distributed in two input channels, namely, the first input channel contains a two-photon state, while another single-photon state is in the second input channel. The excitation probabilities of the two-level system are simulated with different bandwidths of input photons. The influence of the number of input photons and channels on the atomic excitation is concluded.

In this paper, we consider that a quantum linear system is driven by two input channels, which contain a single-photon state and an unfactorizable two-photon state. Quantum entanglement between the two channels is analyzed by deriving the explicit forms of the input and output covariance functions. Simulations are conducted by tunning the coupling strengths between the two input channels and the quantum system, which has certain guiding effect on the design and application of quantum optical experiments.

In this paper, the H ∞ controller design problem is investigated for the discrete-time Markovian jump linear system in the presence of incomplete transition descriptions. A less conservative criterion is presented such that the considered system is stochastically stable with a guaranteed H ∞ performance index by using the properties of the transition probabilities. In addition, a H ∞ state-feedback controller is proposed for the discrete-time Markovian jump system with incomplete transition description via LMI formulation. Finally, some simulation results are given to verify the effectiveness of the presented methods.

Attitude stabilization control via disturbance observer for flexible spacecraft is studied in this paper. First, a model based on Takagi-Sugeno fuzzy system of flexible spacecraft is established. In the case of spacecraft subjected to the external disturbance, the external disturbance and flexible accessories are treated as one disturbance. Furthermore, the lumped disturbance is no longer limited by bounded or zero-order derivatives. Then, a disturbance observer is proposed to estimate it while the gains are determined by the convergence condition of the estimation error. According to the obtained estimate, an integral sliding surface is selected so that the control law is partially composed of PDCC. At last, the effectiveness of the disturbance observer and attitude control law are illustrated by an example of a flexible spacecraft.

Conventional direction of arrival (DOA) estimation methods for known waveform sources have often assumed the noise being Gaussian distribution. However, the impulsive noise are frequently encountered in practice, which may severely degrade their estimation performance. To deal with this problem, we firstly construct a hybrid cost function to obtain the matrix related to unknown DOAs and complex amplitudes in the presence of impulsive noise. Then, by incorporating the majorization-minimization (MM) framework, the cost function is optimized iteratively. Finally, the DOAs and complex amplitudes are estimated via using the inherent relationship of the matrix calculated from the MM step. As demonstrated by simulation results, for strongly impulsive noise, the proposed method has a better estimation performance than many existing methods. Moreover, it can handle both weakly and strongly impulsive cases effectively.

In this paper, we investigate the rational fractions in the framework of conjugate product and establish a division ring. Some conjugate properties on the proposed division ring are obtained, and the similarity and consimilarity properties are investigated.

Based on the standard recursive generalized extended least squares (RGELS) algorithm and the filtering based recursive least squares (F-RLS) algorithm, a new type of filtering based algorithm is presented for ARARMAX model in this paper. The core idea is to transform the ARARMAX model into an ARMAX model by introducing a filter for the input and output data. With such a treatment, the system can be decomposed into two subsystems. Then, a recursive identification algorithm is constructed by replacing immeasurable terms in information vectors with their estimates respectively. A numerical simulation example is employed to show the effectiveness of the proposed algorithm.

In this paper, the convergence characterization of a special implicit iterative algorithm with a tuning parameter for continuous coupled Markov jump Lyapunov matrix equation is investigated. First, a necessary condition for the convergence of the iterative algorithm is given. Then, a necessary and sufficient condition is proposed and the optimal tuning parameter such that the algorithm has the fastest convergence rate is analyzed in two cases according to the distribution of eigenvalues. Finally, a numerical example is given to illustrate the effectiveness of the algorithm and the effects of different tuning parameters.

In this paper, a modified predictor based on the predictive scheme proposed by V. Lechappe is developed to construct a predictor-based controller. It has been shown that this controller can attenuate the unknown additive disturbances for linear time invariant (LTI) systems with a constant and known input delay. Moreover, a comparison with the work from V. Lechappe is presented, which proves that a better attenuation is achieved by utilizing the modified predictor-based controller for time-varying disturbances. And for constant disturbances, the controller based on this modified predictor can also reject the disturbances perfectly. Finally, a numerical example is given to verify these results about constant and time-varying disturbances.

The combination-accelerated stress test for 4 solar cell mini modules is effective for accelerating degradation phenomena observed for those modules exposed outside. The combination test include high fidelity ultraviolet (UV) light exposure, copper accelerated acetic acid salt spray (CASS) test, thermal-cycle (TC) test, and humidity-freeze (HF) test. This test can help us understand whether different materials and process parameters will affect reliability test. In this experiment, the different backsheets, EVA and POE combinations can be found that the power loss < 10%. In addition, it was also observed that the CASS test was not very serious for the mini modules. It indicates that the module without aluminum frame also has good corrosion resistance for CASS testing.

In this paper, the dynamic sliding mode control approach is adopted to design attitude tracking control law for flexible spacecraft. First, the mathematical model of the flexible spacecraft based on attitude errors is constructed. The kinematic equation of the flexible spacecraft is described by error quaternions. Then, by introducing a dynamic switching function, an dynamical sliding mode control law is proposed for the flexible spacecraft to tackle attitude tracking issue. The proposed dynamic sliding mode control law can suppress the chattering raised by the sign function of the traditional sliding mode control law. Finally, a numerical example is employed to verify the effectiveness of the presented control law.

In this paper, we introduce the attempts of our fusion methods during the first DIHARD challenge. To our knowledge, this is the first launch in speaker diarization domain which aims to evaluate the performance of the state-of-the-art system in realistic adverse acoustic environments. Besides speech preprocessing modules including speech denoising and speech activity detection, our attention has been focused on back-end clustering algorithms, especially in system fusion. Consensus clustering is adopt to combine both original speech and denoised speech, for purifying unreliable clusters. Moreover, a score-level fusion is conducted between GMM-UBM-based i-vector and CNN-based i-vector. Finally, our system achieves diarization error rates (DERs) of 36.05% on the evaluation sets, which is the second place in the DIHARD challenge.

Apparent quality of concrete is one of the important indexes of engineering quality assessment, occupying a very important position in the quality of engineering. As an important evaluation index of the apparent quality of concrete, the detection of cracks is particularly important. The traditional detection method mainly relies on the inspection personnel to carry out on-site inspection, which has low efficiency and large environmental limitations. In this paper, a method of crack image recognition based on convolution neural network is proposed, which can realize the automatic recognition of concrete surface cracks, and the recognition accuracy rate is as high as 93.7{\%.

To solve the active vibration isolation problem for Stewart platforms, an adaptive backstepping sliding mode controller is designed in this paper. Firstly, the dynamic model is constructed for the Stewart platform. Then, the dynamics is decoupled by using a simultaneous diagonization, and an adaptive backstepping sliding mode controller is designed such that the closed-loop system is stable and the system error can converge to zero. Finally, the effectiveness of the proposed method is verified by a simulation example.

In this paper, by introducing a tuning parameter, an explicit iterative algorithm is constructed for solving the continuous Lyapunov matrix equations associated with Ito stochastic systems. A necessary and sufficient condition is provided to guarantee the convergence of the proposed algorithm by using the Kronecker product. Moreover, some easily verifiable convergence results are developed for the proposed algorithm based on the obtained necessary and sufficient condition. In addition, a method is given for the selection of the optimal tuning parameter. Finally, an example is applied to verify the effectiveness of the presented algorithm.

In this work, we proposed the different laserdoped selective emitters (LDSE) parameters for the industrial mono silicon solar cell applied to screen-printed passivated emitter and rear cell (PERC) structure. We compared the 156.75 × 156.75 mm2 p-type mono silicon PERC process with LDSE process and without LDSE process. It can be found that the different laser parameters induced different sheet resistance. Laser induced damage is investigated for different laser energies with SEM images. The successful LDSE technology PERC solar cells were also increased an absolute gain of 0.15 ~ 0.28 % in average efficiency compared with the standard PERC solar cells. Moreover, it is worthy to note that many previous references have shown that LDSE technology could enhance the efficiency which can be increased by more than 0.4%. In fact, with the improvement of silver paste technology, the LDSE technology can enhance the performance should be only about 0.1 ~ 0.2 %.

Protocol reverse analysis is a research hotspot in the field of information confrontation and network security. Inferring the format of unknown binary protocols from network traces is one challenge task in protocol reverse engineering. In this study, a protocol reverse method for binary protocol which combined with hierarchical clustering and probability alignment is proposed. First, the method classifies the frames with same format into one set based on hierarchical clustering. Second, probability alignment algorithm is adopted to align the corresponding fields in protocol frame accurately. Finally, according to the characteristics of the statistical data, the boundary of the adjacent fields in the protocol is determined, and the protocol format is extracted according to the field boundary. The experimental results show that the method can effectively identify the format specification of the binary protocol and the semantic specification of some fields.

A bias compensation based least-squares algorithm is proposed for the parameter estimation of multi-input single-output system in the presence of input and output white noises. It is shown that the bias term is induced by the variances of input and output noises. Therefore, an efficient method which uses the observed input and output data directly is developed in this paper to estimate the unknown variances of white noises. The proposed bias compensation based least-squares algorithm can be established from the combination of the recursive least-squares estimation algorithm and white noise variances estimation algorithm. The effectiveness of the proposed algorithm is both analyzed theoretically and verified by a simulation example.

In this paper, we consider the problem of attitude control for flexible satellites based on the quaternion model with unknown bounded external disturbance. A sliding mode control law is designed to attenuate the effects of the disturbance for the closed-loop system. In addition, an observer-based sliding mode control law is proposed when the flexible modal variables of the flexible satellites are not available. Finally, a numerical example is given to show the proposed controllers.

In this paper, an implicit iterative algorithm is proposed to obtain the unique positive definite solution of the continuous algebraic Riccati matrix equation. In this proposed algorithm, there exists a tuning parameter which can be appropriately chosen such that the algorithm achieves better convergence performance. A sufficient condition is given for the convergence of the proposed algorithm. Moreover, an approach is also provided to choose an appropriate initial condition. Finally, a numerical example is employed to show the effectiveness of the proposed algorithm.