基于转移学习的中文命名实体识别

中文命名实体识别在多个重要领域有广泛的运用,提出一种基于转移学习的算法进行中文命名实体识别,旨在提高识别的准确率和召回率。基于转移学习算法的中心思想是开始以一些简单的结论应用于问题,然后在每个步骤应用转换,选择出每次转换的最优结论再次应用于问题,当选择的转换在足够的空间内不再修改数据时算法停止。提出算法的规则模板和约束文件的获取方法,形成一个完整的用于中文命名实体识别的模型,并利用该模型进行实验,获得了较好的结果。

基于转移学习的命名实体挖掘技术

研究了针对大规模查询日志中丰富的命名实体的挖掘技术,通过利用Wikipedia数据,结合转移学习方法构建目标类别的分类器.该技术很好地利用了监督学习的优越性能以提高查询日志中命名实体挖掘的准确性,同时也解决了监督学习方法中大规模标注的问题.实验结果表明,基于转移学习的命名实体挖掘技术具有优越的命名实体挖掘性能.

基于转移学习的小样本数据深度学习研究

卷积神经网络的深度学习在图像识别领域取得了巨大的成功,但是训练一个深度学习网络需要大量的数据样本。在实际工作中,很难得到大量的训练样本,在数据集有限的情况下,容易过度拟合。针对这一问题,设计了一种基于转移学习的深度卷积神经网络来解决小样本数据集的问题。采用数据扩充的方法来扩大样本数据集的数量,利用转移学习将训练好的网络(CNN)从大样本数据集中转移到的小样本数据集中进行二次训练,使用全局平均池而不是全连接层来训练网络,并利用Soft max进行分类。该方法解决了深度学习中样本数据集小的问题,提高了操作效率。实验结果表明,该方法对小样本数据集的分类具有较高的识别率。

基于转移学习改进低开销指纹的室内定位算法研究

基于指纹定位算法,为了降低在离线状态目标域的站点测量开销而提高定位的准确性,提出一种通过减少离线训练开销,但是不影响定位精度的基于转移学习(TL)框架的室内定位算法。其基本原理是,根据源域的知识转移重构目标域中的数据分布,使得属于同一个簇的数据在逻辑上彼此接近,而剩余数据则彼此分开。基于TL的框架由2部分组成:度量学习和度量转移,分别用于从源域学习距离度量,并分别为目标域确定最合适的度量。最后利用仿真实验结果证明了基于TL框架定位系统的有效性。

谈学习转移及其在俄语二外教学中的应用

指出了掌握英语的学习者学习俄语过程中会出现英语语言知识对俄语学习的影响。从心理语言学及学习心理学的角度阐述了学习转移的分类及其产生的原因 ,分析了影响转移的因素 。

组织交互渠道、关系质量及学习意愿对中国-东盟跨国公司技术转移绩效的影响

基于东盟子公司员工的中微观视角,通过问卷调查,采用PLS-SEM结构方程模型,分析和检验中国-东盟跨国母公司与子公司间的交互渠道、关系质量、学习意愿3个因素对技术转移绩效的影响机理和路径。研究结果发现,在中国-东盟跨国公司内部,丰富和便捷的交互渠道对技术转移绩效具有显著直接正向影响,同时通过关系质量、学习意愿对技术转移绩效产生间接正向影响。关系质量不能直接促使技术转移绩效提升,但可以通过学习意愿的中介作用间接影响技术转移绩效。学习意愿对技术转移绩效具有显著直接正向影响。

模态桥传递学习方法在医学图像分类中的应用

为减轻医疗领域中标签数据不足的问题。本文提出了一种基于转移学习的医疗图像分类的新方法。提出了一种桥梁数据库与目标数据库采用相同的医学成像采集方式的模式桥传输学习方法,通过学习从源到桥和从桥到目标的投影函数,可以减轻源(例如自然图像)和目标(例如X射线图像)之间的区域差异。实验结果表明,与各种传输学习方法相比,所提出的方法即使对于少量标记的目标医学图像也能够实现高分类性能。

创业学习转移:模型构建及机制分析

创业学习已有研究成果多关注创业者学习,且质性研究居多,对员工、群体和组织等其他学习主体的创业学习机制以及不同主体之间的学习转移机制缺少关注。为弥补这一研究不足,本文从创业学习转移这一全新视角和崭新话题推进创业学习理论研究。本文构建了创业学习概念体系的5W1H模型,从何时学习,为什么学习,由谁学习,从哪学习,学习什么,以及如何学习六个方面系统完善创业学习的概念体系;构建了创业学习转移的理论模型;分析了创业者、员工、群体和组织等不同层面学习主体的创业学习机制;构建了创业者到员工的创业学习转移,员工到群体的创业学习转移,以及群体到组织的创业学习转移的理论模型,揭示了不同层面创业学习主体之间创业学习转移的路径与机制。

对学习中心转移的科学把握

评说理论界关于学习中心转移观念的不同认识，指出学习中心论（及其相应的教学模式）和教学中心论（及其相应的教学模式）都存在片面性。扬弃它们的片面解释，辩证地把握学习的社会性、主体性，是学习中心转移的要旨，是教育科学发展的大趋势。

Action Recognition from Videos with Complex Background via Transfer Learning

Classifier learning methods commonly assume that the training data and the testing data are drawn from the same underlying distribution. However, in many practical situations, this assumption is violated. One example is the practical action videos with complex background and the universal human action databases of Kangliga Tekniska Hogskolan （KTH）. When training data are very scarce, supervised learning is difficult. However, it will cost lots of human and material resources to establish a labeled video set which includes a large amount of videos with complex backgrounds. In this paper, we propose an action recognition framework which uses transfer boosting learning algorithm. By using this algorithm, we can train an action recognition model fitting for most practical situations just relaying on the universal action video dataset and a tiny set of action videos with complex background. And the experiment results show that the performance is improved.

Numerical Study on Heat Transfer and Flow Characteristics for Laminar Flow in a Circular Tube with Swirl Generators

This study investigated the heat transfer and flow characteristics of one kind of swirlgenerator in a circular heat exchanger tube through a numericalsimulation. The swirlflow induced by this type of swirlgenerator can obtain a high heat transfer rate with minimalpressure drop penalty. The simulations were carried out to understand the physicalbehavior of this kind of mesoscale heat enhancement component. By visualizing the heat transfer and flow characteristics, it is found that the swirlflow is induced by swirlgenerator in the circular tube couples with the impinging jet effect. After passing through the swirlgenerator, the localfriction factor of liquid can quickly return to lower levelmore quickly, while the localNusselt number maintains higher values for a distance; thus, the evaluation criterion of localperformance is improved. Single-factor optimization is used for three geometric parameters, i.e., the angle of swirlgenerator(25o, 45o, and 60o), the length of swirlgenerator(0.005, 0.01, and 0.02 m), and the center rod radius(1, 2, and 3 mm). The optimum parameters of the swirlgenerator for laminar flow of air in a circular tube are obtained, which should be 60o, 0.005 m, and 3 mm, respectively.

Engine Noise Prediction by Using Multi-body Simulation

Abstract： This paper presents a coupled multi-body and FEM （finite element method）-BEM （boundary element method） methodology used to carry out a comprehensive NVH （noise, vibration and harshness） investigation of a four-cylinder internal combustion engine prototype. Firstly, a MBDS （multi-body dynamic simulation） of the internal combustion engine has been carried out, at a defined operating condition, in order to determine the excitation force of the powertrain exciting the cylinder block. In this way, the dynamics of the engine powertrain have been described taking into account both the effects of the gas forces of the combustion process and the inertia forces of the moving parts. Afterwards, the cylinder block excitation forces have been used to evaluate the engine block vibrations and to predict the external noise radiated with both the well-known ATV （acoustic transfer vectors） and MATV （modal acoustic transfer vectors） methodologies at a distance of 1 m from the engine, according to the standard ISO 3744. The dynamics of the engine powertrain and its vibro-acoustic behaviour have been described using LMS （learning management system） Engineering Innovation Virtual.Lab tools.

合法性危机下跨国公司人力资源本土化策略及其动态演进研究——基于爱菊粮油集团的案例研究

跨国公司人力资源本土化策略是解决合法性压力的有效途径,但新兴经济体跨国公司呈现出与发达国家跨国公司不同的本土化策略及路径。本文以中国企业爱菊粮油工业集团有限公司为研究对象,通过对案例数据进行分析,探索了新兴经济体跨国公司不同的人力资源本土化策略类型及其演化路径。研究发现:(1)为了化解合法性压力问题,新兴经济体跨国公司会采取适应与融合两种不同的人力资源本土化策略。其中,适应策略适用于缓解外部合法性危机,而融合策略则兼顾了内外部合法性问题。(2)随着外部合法性危机向内部合法性危机转化,人力资源本土化从适应策略向融合策略演进,而“桥梁”人才资源的规划利用是解决内外部合法性问题的关键。(3)知识学习嵌入在人力资源本土化过程中,由单向知识学习向双向知识学习转化。本文丰富了跨国公司人力资源本土化策略、合法性压力化解等相关理论,对新型经济体跨国公司的人力资源本土化战略选择有重要的借鉴价值。

Comparative Study of Convective Heat Transfer Performance of Steam and Air Flow in Rib Roughened Channels

A comparative experimental study of heat transfer characteristics of steam and air flow in rectangular channels roughened with parallel ribs was conducted by using an infrared camera. Effects of Reynolds numbers and rib angles on the steam and air convective heat transfer have been obtained and compared with each other for the Reynolds number from about 4,000 to 15,000. For all the ribbed channels the rib pitch to height ratio(p/e) is 10, and the rib height to the channel hydraulic diameter ratio is 0.078, while the rib angles are varied from 90° to 45°.Based on experimental results, it can be found that, even though the heat transfer distributions of steam and air flow in the ribbed channels are similar to each other, the steam flow can obtain higher convective heat transfer enhancement capability, and the heat transfer enhancement of both the steam and air becomes greater with the rib angle deceasing from 90° to 45°. At Reynolds number of about 12,000, the area-averaged Nusselt numbers of the steam flow is about 13.9%, 14.2%, 19.9% and 23.9% higher than those of the air flow for the rib angles of 90°,75°, 60° and 45° respectively. With the experimental results the correlations for Nusselt number in terms of Reynolds number and rib angle for the steam and air flow in the ribbed channels were developed respectively.

Experimental Study on Flow Boiling of Deionized Water in a Horizontal Long Small Channel

In this paper,an experimental investigation on the flow boiling heat transfer in a horizontal long mini-channel was carried out. The mini-channel was with 2 mm wide and 1 mm deep and 900 mm long. The material of the mini-channel was stainless. The working fluid was deionized water. The experiments were conducted with the conditions of inlet pressure in the range of 0.2~0.5 MPa, mass flux in the range of 196.57-548.96 kg/m2 s, and the outlet vapor quality in the range of 0.2 to 1. The heat flux was in the range of 292.86 kW/m2 to 788.48 kW/m2,respectively. The influences of mass flux and heat flux were studied. At a certain mass flow rate, the local heat transfer coefficient increased with the increase of the heat flux. If dry-out occurred in the mini-channel, the heat transfer coefficient decreased. At the same heat flux, the local heat transfer coefficient would depend on the mass flux. It would increase with the mass flux in a certain range,and then decrease if the mass flux was beyond this range. Experimental data were compared with the results of previous studies. Flow visualization and measurements were conducted to identify flow regime transitions. Results showed that there were eight different kinds of flow patterns occurring during the flow boiling. It was found that flow pattern had a significant effect on heat transfer.

Study of Variable Turbulent Prandtl Number Model for Heat Transfer to Supercritical Fluids in Vertical Tubes

In order to improve the prediction performance of the numerical simulations for heat transfer of supercritical pressure fluids, a variable turbulent Prandtl number(Pr_t) model for vertical upward flow at supercritical pressures was developed in this study. The effects of Pr_t on the numerical simulation were analyzed, especially for the heat transfer deterioration conditions. Based on the analyses, the turbulent Prandtl number was modeled as a function of the turbulent viscosity ratio and molecular Prandtl number. The model was evaluated using experimental heat transfer data of CO_2, water and Freon. The wall temperatures, including the heat transfer deterioration cases, were more accurately predicted by this model than by traditional numerical calculations with a constant Pr_t. By analyzing the predicted results with and without the variable Pr_t model, it was found that the predicted velocity distribution and turbulent mixing characteristics with the variable Pr_t model are quite different from that predicted by a constant Pr_t. When heat transfer deterioration occurs, the radial velocity profile deviates from the log-law profile and the restrained turbulent mixing then leads to the deteriorated heat transfer.