预训练模型辅助的后门样本自过滤防御方法

深度神经网络由于其出色的性能,被广泛地部署在各种环境下执行不同的任务,与此同时它的安全性变得越来越重要。近年来,后门攻击作为一种新型的攻击方式,对用户构成严重威胁。在训练阶段,攻击者对少量样本添加特定后门模式并标记为目标类以学习后门模型。后门模型可以以很高的概率将加入后门模式的测试样本识别为目标类,同时不影响正常样本的识别。用户通常无法掌握后门的先验信息,因此很难察觉后门攻击的存在。该文提出一种预训练模型辅助的后门样本自过滤方法,以防御后门攻击,包括目标类检测与后门样本自过滤两个部分。在第一部分,利用预训练模型提取样本特征,采用k近邻算法进行目标类检测;在第二部分,从非目标类样本中学习部分分类模型,之后多次执行“后门样本过滤”与“模型学习”的交替计算,在较好过滤后门样本的同时,也得到了完整的良性模型。

新型自过滤检查井开发与智能化应用探索

自过滤的塑料检查井,可以有效防止地下管网的堵塞。自过滤检查井包含过滤网井筒、井座和井盖。井筒表面设置有网状加强筋,能有效提高产品强度,很好地保护塑料检查井不被震破或者压迫变形,提高塑料检查井的抗震效果,增加了使用寿命。过滤网可以很好地防止杂物进入检查井,避免连接检查井的管道堵塞。进一步通过对管网中检查井智能化方案,通过数据无线传输,实现对智能管网的预报预警。

煤气化装置在线分析仪表预处理系统优化改造

在线分析仪在石油化工生产中有着广泛的应用,特别是在煤气化装置中,其在线分析预处理系统承担着极为关键的作用。由于煤气化装置的工况复杂恶劣,在线分析仪表系统的应用成为瓶颈。详细介绍了煤气化装置在线分析预处理系统自投入使用以来,所进行的摸索与优化改造。经过优化改造,在线分析预处理系统完全克服了煤气化粗合成气粉尘大、水量多、温度高的不足,效果显著,保证了分析仪系统长期安全、稳定、可靠的运行。

Numerical simulation for determining three zones in the goaf at a fully-mechanized coal face

According to the theory of ventilation network, a model for a filtration flow field in goaf was built and simulation soft- ware for determining the three zones in goaf has been developed. This software uses no-gap-connection between Visual Basic (VB) and Excel to exchange data, uses Component Object Model (COM) component of MATLAB to plot charts of the three zones and to export the corresponding coordinates of the isolines. An example shows that this software is convenient and simple. By using it, the three zones can be easily determined. The software is convenient in studies and analyses of the three zones in goaf.

Electronic Transport Properties of Spin-Crossover Magnet Fe（Ⅱ）-N4S2 Complexes

Spin-crossover （SCO） magnets can act as one of the most possible building blocks in molec- ular spintronics due to their magnetic bistability between the high-spin （HS） and low-spin （LS） states. Here, the electronic structures and transport properties through SCO magnet Fe（II）-N4S2 complexes sandwiched between gold electrodes are explored by performing exten- sive density functional theory calculations combined with non-equilibrium Green＇s function formalism. The optimized Fe-N and Fe-S distances and predicted magnetic moment of the SCO magnet Fe（II）-N4S2 complexes agree well with the experimental results. The reversed spin transition between the HS and LS states can be realized by visible light irradiation according to the estimated SCO energy barriers. Based on the obtained transport results, we observe nearly perfect spin-filtering effect in this SCO magnet Fe（II）-N4S2 junction with the HS state, and the corresponding current under small bias voltage is mainly contributed by the spin-down electrons, which is obviously larger than that of the LS case. Clearly, these theoretical findings suggest that SCO magnet Fe（II）-N4S2 complexes hold potential applications in molecular spintronies.

Negative Differential Resistance and Spin-Filtering Effects in Zigzag Graphene Nanoribbons with Nitrogen-Vacancy Defects

We explore the electronic and transport properties of zigzag graphene nanoribbons （GNRs） with nitrogen-vacancy defects by performing fully self-consistent spin-polarized density functional theory calculations combined with non-equilibrium Green＇s function technique. We observe robust negative di erential resistance （NDR） effect in all examined molecular junctions. Through analyzing the calculated electronic structures and the bias-dependent transmission coefficients, we find that the narrow density of states of electrodes and the bias-dependent effective coupling between the central molecular orbitals and the electrode subbands are responsible for the observed NDR phenomenon. In addition, the obvious di erence of the transmission spectra of two spin channels is observed in some bias ranges, which leads to the near perfect spin-filtering effect. These theoretical findings imply that GNRs with nitrogenvacancy defects hold great potential for building molecular devices.

Adaptive bands filter bank optimized by genetic algorithm for robust speech recognition system

Perceptual auditory filter banks such as Bark-scale filter bank are widely used as front-end processing in speech recognition systems.However,the problem of the design of optimized filter banks that provide higher accuracy in recognition tasks is still open.Owing to spectral analysis in feature extraction,an adaptive bands filter bank (ABFB) is presented.The design adopts flexible bandwidths and center frequencies for the frequency responses of the filters and utilizes genetic algorithm (GA) to optimize the design parameters.The optimization process is realized by combining the front-end filter bank with the back-end recognition network in the performance evaluation loop.The deployment of ABFB together with zero-crossing peak amplitude (ZCPA) feature as a front process for radial basis function (RBF) system shows significant improvement in robustness compared with the Bark-scale filter bank.In ABFB,several sub-bands are still more concentrated toward lower frequency but their exact locations are determined by the performance rather than the perceptual criteria.For the ease of optimization,only symmetrical bands are considered here,which still provide satisfactory results.

Adaptive Gaussian Noise Image Removal Algorithm Using Filtering-Based Noise Estimation

This paper proposes a spatially denoising algorithm using filtering-based noise estimation for an image corrupted by Gaussian noise.The proposed algorithm consists of two stages:estimation and elimination of noise density.To adaptively deal with variety of the noise amount,a noisy input image is firstly filtered by a lowpass filter.Standard deviation of the noise is computed from different images between the noisy input and its filtered image.In addition,a modified Gaussian noise removal filter based on the local statistics such as local weighted mean,local weighted activity and local maximum is used to control the degree of noise suppression.Experiments show the effectiveness of the proposed algorithm.

Fault detection and identification for dead reckoning system of mobile robot based on fuzzy logic particle filter

To deal with fault detection and diagnosis with incomplete model for dead reckoning system of mobile robot,an integrative framework of particle filter detection and fuzzy logic diagnosis was devised.Firstly,an adaptive fault space is designed for recognizing both known faults and unknown faults,in corresponding modes of modeled and model-free.Secondly,the particle filter is utilized to diagnose the modeled faults and detect model-free fault according to the low particle weight and reliability.Especially,the proposed fuzzy logic diagnosis can further analyze model-free modes and identify some soft faults in unknown fault space.The MORCS-1 experimental results show that the fuzzy diagnosis particle filter(FDPF) combinational framework improves fault detection and identification completeness.Specifically speaking,FDPF is feasible to diagnose the modeled faults in known space.Furthermore,the types of model-free soft faults can also be further identified and diagnosed in unknown fault space.

A new improvement of TCP based on adaptive filtering

On the basis of studying the standard TCP retransmission mechanism, this paper proposes a method to adopt the theory of adaptive filtering in the field for the estimation of round trip time （RTT）. Then this method is embedded in TCP for estimation of RTT, and defined as RTT-AF TCP. In the schemes with Only-TCP flow and with TCP and UDP flow fighting for the link, the experiment results were made and analyzed by using the standard TCP and the RTT-AF TCP. The experimental results show higher throughput and lower loss rate of packets by using RTT-AF TCP for data transmission, and achieve excellent effect.

浅析教辅数字出版广告策略

据相关数据显示,如今出版社的图书销售有很大一部分的盈利来自教辅书的销售,教辅数字出版将成为未来教辅类出版竞争的新领域。本文从我国现行的教辅数字化出版模式开始分析,对教辅数字出版的广告形态与传播模式进行讨论,进而引出我国教辅数字出版行业自过滤带来的发展以及广告营销中面临的挑战。

Alignment control for a long span urban rail-transit cable-stayed bridge considering dynamic train loads

In this paper, bridge alignment control with considering dynamic train loads was experimentally and theoretically investigated.Analytical process of bridge dynamics and the self-adaptive Kalman filter bridge alignment control method with considering the dynamic train loads were briefly introduced. The static measurement, the dynamic test, the field alignment measurement as well as the finite element analysis(FEA) of the second longest rail transit cable-stayed bridge in the world were carried out.Based on the results, the train dynamic load effect on the bridge alignment was obtained quantitatively. Subsequently, alignment control of the rail transit bridge with considering this effect using a self-adaptive Kalman filter method was analyzed. The results show that:(a) the dynamic train loads have effects on alignment control of the bridge and therefore cannot be neglected;(b) the self-adaptive Kalman filter method is applicable and reliable for alignment control of bridges during construction. The analytical method and whole process contribute to develop a related specification and further engineering applications.

Spin-Filtering Transport in Double Parallel Quantum Wires on a Graphene Sheet

We theoretically investigate the spin filtering transport of double parallel quantum wires(QWs) side-coupled to a grapheme sheet and sandwiched between two ferromagnetic(FM) leads.The dependences of the wire-graphene coupling strength,wire-wire coupling strength,as well as the spin polarization of the ferromagnetic leads are studied.It is found that the wire-graphene coupling strength tends to reduce the current and the wire-wire coupling strength can first reinforce and then decrease the current.The spin polarization strength has an enhanced(identical) effect on the current under the parallel(anti-parallel) alignment of the FM leads,which gives rise to an obvious spin-filter and tunnel magnetoresistance(TMR) effect.Our results suggest that such a theoretical model can stimulate some experimental investigations about the spin-filter devices.

An improved adaptive Sage filter with applications in GEO orbit determination and GPS kinematic positioning

The shortcomings of an adaptive Sage filter are analyzed in this paper.An improved adaptive Sage filter is developed by using a weighted average quadratic form of the historical residuals of observations and predicted states to evaluate the covariance matrices of observations and dynamic model errors at the present epoch.The weight function is constructed based on the variances of observational residuals or predicted state residuals and the space distance between the previous and the present epoch.In order to balance the contributions of the measurements and the dynamic model information,an adaptive factor is applied by using a two-segment function and predicted state discrepancy statistics.Two applications,orbit determination of a maneuvered GEO satellite and GPS kinematic positioning,are conducted to verify the performance of the proposed method.

Simple Adaptive Filtering Scheme to Improve Measurement Accuracy of Gyroscope on Angular Motion Base

The objective of this work is to improve the measurement accuracy of a gyroscope on a angular motion base with a simple adaptive filter scheme.Two main topics are highlighted in this work.The first topic is to show building a dual-process model employed for the conventional Kalman filter.The second topic is to show developing a modified noise adaptive algorithm when measurement noise and process noise are unknown.The experimental results are presented to show that the simple adaptive filtering scheme outperforms the other conventional scheme in this paper in terms of noise reduction.

Filtering and tracking with trinion-valued adaptive algorithms

A new model for three-dimensional processes based on the trinion algebra is introduced for the first time.Compared to the pure quaternion model, the trinion model is more compact and computationally more efficient,while having similar or comparable performance in terms of adaptive linear filtering. Moreover, the trinion model can effectively represent the general relationship of state evolution in Kalman filtering, where the pure quaternion model fails. Simulations on real-world wind recordings and synthetic data sets are provided to demonstrate the potential of this new modeling method.

ADAPTIVE TRACKING CONTROL FOR ACTUATOR FAILURE COMPENSATION BASED ON MT-FILTERS

This paper considers the problem of actuator failure compensation for a class of nonlinearsystems in the form of output-feedback.MT-filters are exploited to estimate the unmeasured states.The backsteppping technique is used to construct the controller recursively.The proposed compensationscheme can grantee the boundedness of all the closed-loop signals and asymptotic output tracking.Thetheoretical results are illustrated by a simulation example.