迁移组概率学习机

基于组概率的学习方法因其能够很好地保护数据的隐私性而成为近年来机器学习领域的研究热点.已有的组概率学习方法虽然取得了一定的效果,但是在模型训练时仅考虑单一的场景信息,如果在当前领域所采集的数据信息有限,则在当前领域下建立的分类模型泛化能力较差.针对此问题,提出了一种基于组概率和结构风险最小化模型的迁移组概率学习机(TGPLM).该方法通过构造领域相似距离项来引入历史领域的先验知识,提出了针对类标签保护数据的增强型分类器优化目标学习准则,以期在有效利用当前领域数据类标签组概率信息的同时借鉴历史领域相关知识来指导当前领域下的学习任务.基于模拟、UCI及PIE人脸等数据集上的实验结果表明,本文所提之方法是有效的.

基于不确定Voronoi图的概率组最近邻查询

为了有效解决不确定对象的组最近邻查询问题,对不确定Voronoi图进行分析,提出单核不确定Voronoi区域和多核不确定Voronoi区域的概念,研究查询组对象的分布特点,利用不确定Voronoi图的性质,给出判断概率组最近邻的理论方法,并提出基于不确定Voronoi图的概率组最近邻查询算法,实验结果表明,算法的查询时间明显减少,具有较好的查询性能。

基于概率和权重信息的二元组概率权重研究

由于现存的三种概率权重表示方法在运算过程中会造成一定程度的信息丢失,所以本文结合概率和权重信息提出二元组概率权重的概念。为了不失一般性,介绍了实值二元组,并借助实值二元组的基本运算、运算性质和排序函数来说明二元组概率权重的相关概念。最后,基于二元组概率权重提出了二元组概率权重有序平均(2TPWOA)算子,并通过具体实例和比较分析说明了二元组概率权重的有效性和实用性。

Statistical Characteristics of Complete Tobacco Necrosis Virus Genome and Clustering

[Objective] The study aimed to extract the statistical features of complete tobacco necrosis virus genome and conduct clustering analysis.[Method] On the base sequence of complete tobacco necrosis virus genome,by using three base groups which was composed of each base and its subsequent two bases,a new sequence S was arranged;the probability of all 64 kinds of three base groups occurred on S was calculated,a 64-dimensional vector L was obtained;L-vector of each genome was compared,8 three base groups were obtained,their probabilities have significant differences.[Result] The emergence probability of 8 three base groups （CGT;CAA;TAC;TAA;AGG;ATG;ATA;AAG） has significant association with genetic variation of tobacco necrosis virus genome;according to genetic variation result,4 different sources tobacco necrosis virus complete genome formed 2 major categories.[Conclusion] The study will provide theoretical basis for the control of tobacco necrosis virus.

Luminescent Efficiency in Single-layer Organic Electrophosphorescent Devices

Based on the charge injection and recombination processes and the triplet-triplet annihilation process, a model to calculate the electroluminescent(EL) efficiency is presented. The influences of the applied electric field on the injection efficiency, recombination efficiency and electroluminescent efficiency are discussed. It is found that: (1) The injection efficiency is increasing while the recombination efficiency is decreasing with the applied electric field increasing. (2) The EL efficiency is enhanced at low electric field slowly but is decreasing at high electric field with the increase of applied voltage. (3) The EL efficiency is decreasing with the increase of the host-guest molecular distance (R). So, it is concluded that the EL efficiency in single-layer organic electrophosphorescent devices is dominated by injection efficiency at lower electric field and recombination efficiency at higher electric field.

Dynamic probabilistic analysis of stress and deformation for bladed disk assemblies of aeroengine

In order to describe and control the stress distribution and total deformation of bladed disk assemblies used in the aeroengine, a highly efficient and precise method of probabilistic analysis which is called extremum response surface method(ERSM) is produced based on the previous deterministic analysis results with the finite element model(FEM). In this work, many key nonlinear factors, such as the dynamic feature of the temperature load, the centrifugal force and the boundary conditions, are taken into consideration for the model. The changing patterns with time of bladed disk assemblies about stress distribution and total deformation are obtained during the deterministic analysis, and at the same time, the largest deformation and stress nodes of bladed disk assemblies are found and taken as input target of probabilistic analysis in a scientific and reasonable way. Not only their reliability, historical sample, extreme response surface(ERS) and the cumulative probability distribution function but also their sensitivity and effect probability are obtained. Main factors affecting stress distribution and total deformation of bladed disk assemblies are investigated through the sensitivity analysis of the model. Finally, compared with the response surface method(RSM) and the Monte Carlo simulation(MCS), the results show that this new approach is effective.

A Research and Design on Application of Genetic Algorithm

Traditional algorithms of intelligent test paper generation have some shortcomings of slow generating speed, low success probability and poor generating quality. Genetic Algorithm is a random optimization-searching algorithm based on probability. In this paper, based on the user＇ s request of intelligent test paper generation, we establish the mathematical model and object function, and give some improvements in Simple GA for test paper generation. Based on the algorithm, a automatic generating test paper system is designed and implemented.

Test for Investor Rationality for Companies Listed at the Nairobi Stock Exchange

Investors have traditionally been viewed as economically rational individuals who make decisions based on all available information. They have been assumed to use probability functions to arrive at the most optimal decision. More recent studies propose that investors are irrational and systematically overreact to good and bad information events. The concept of the rational investor has been supported by among others Efficient Market Hypothesis and Modem Portfolio Theory. Other studies opposed to the notion of rational investors have identified psychological biases that influence decision making process of an investor, and leading them to make irrational decisions. Several anomalies have been identified that deviate from rational behavior. The objective of this paper was to test for investor rationality for companies listed at the Nairobi Stock Exchange. This paper tested overreaction by investors to news and performance of companies listed at the Nairobi Stock Market as an anomaly that has been proven in other markets. The test involved forming companies into two portfolios, one of extreme good performers and the other of extreme poor performers during the base year. Performance of these portfolios was analyzed for a nine year period from the year of portfolio formation. The results are consistent with the notion of overreaction, showing that investors overreact to both good and bad news. Over the study period the loser portfolio outperformed the winner portfolio by about 35.92%. This confrere that investors are irrational and make decisions based on some biases.

Evaluation of Loss of Load Probability and Expected Energy Generation in Multi-area Interconnected Systems with Wind Generating Units

Due to randomness of wind generator＇s availability, power system planners have big concern on EEG （expected energy generation） and system reliability of power system with wind generators. This paper presents a methodology to evaluate the EEG as well as overall LOLP （loss of load probability）, which is an index for system reliability of multi-area interconnected systems with wind generators, as well as conventional fossil fuel based generating units. The proposed model is also capable of tracking the energy export incorporating the multi-state probability model for wind generator which output varies with time and season.

Detection Probability Analysis of Space-Based AIS Signal by Using Geometrical Model

To analyze the detection probability of shipbome AIS （automatic identification system） signal from space, a mathematical model dependent upon three factors of message collision avoidance, power of signal received by satellite and interference ratio of signal received is presented in the paper. The altitude and footprint area of the AIS satellite are discussed to overcome the collision of messages transmitted in the different time slots fxom different SOTDMA （self organizing time division multiple access） cell areas, but arrive at the same time slot due to the different signal path lengths. The simulated result shows that compared to the normal LEO （low ear~ orbit） satellite system, on average the maximum signal coverage area and the maximum FOV （field of view） of the AIS satellite system are reduced by 74% and 38%. The majority of power of signal transmitted fi＇om shipborne 12W-power AIS transmitters located within the maximum signal coverage area may be received with the sufficient margin of power of signal by the LEO satellite, but the space-based AIS system generally suffers from the insufficient CIR （carrier to co-channel interference ratio） of signal received since around 95% pairs of message simultaneously received by satellites may not be correctly decoded. The insufficient CIR of signal received is the bottleneck for the high message detection probability. Therefore, the measure of separating the collision messages should be further taken by the space-based AIS system to increase the detection probability.

Combinatorial and Statistical Applications of Generalized Stirling Numbers

Here illustrated are certain combinatorial and statistical applications of the generalized Stirling numbers with integer parameters.

Probability density function solution to nonlinear ship roll motion excited by external Poisson white noise

The stationary probability density function (PDF) solution to nonlinear ship roll motion excited by Poisson white noise is analyzed. Subjected to such random excitation, the joint PDF solution to the roll angle and angular velocity is governed by the generalized Fokker-Planck-Kolmogorov (FPK) equation. To solve this equation, the exponential-polynomial closure (EPC) method is adopted. With the EPC method, the PDF solution is assumed to be an exponential-polynomial function of state variables. Special measure is taken such that the generalized FPK equation is satisfied in the average sense of integration with the assumed PDF. The problem of determining the unknown parameters in the approximate PDF finally results in solving simultaneous nonlinear algebraic equations. Both slight and high nonlinearities are considered in the illustrative examples. The analysis shows that when a second-order polynomial is taken, the result of the EPC method is the same as the one given by the equivalent linearization (EQL) method. The EQL results differ significantly from the simulated results in the case of high nonlinearity. When a fourth-order or sixth-order polynomial is taken, the results of the EPC method agree well with the simulated ones, especially in the tail regions of the PDF. This agreement is observed in the cases of both slight and high nonlinearities.

Uncertainty Characterization in Remotely Sensed Land Cover Information

Uncertainty characterization has become increasingly recognized as an integral component in thematic mapping based on remotely sensed imagery, and descriptors such as percent correctly classified pixels (PCC) and Kappa coefficients of agreement have been devised as thematic accuracy metrics. However, such spatially averaged measures about accuracy neither offer hints about spatial variation in misclassification, nor are useful for quantifying error margins in derivatives, such as the areal extents of different land cover types and the land cover change statistics. Such limitations originate from the deficiency that spatial dependency is not accommodated in the conventional methods for error analysis. Geostatistics provides a good framework for uncertainty characterization in land cover information. Methods for predicting and propagating misclassification will be described on the basis of indicator samples and covariates, such as spectrally derived posteriori probabilities. An experiment using simulated datasets was carried out to quantify the error in land cover change derived from postclassification comparison. It was found that significant biases result from applying joint probability rules assuming temporal independence between misclassifications across time, thus emphasizing the need for the stochastic simulation in error modeling. Further investigations, incorporating indicators and probabilistic data for mapping and propagating misclassification, are anticipated.