基于出行模式的注意力机制可解释性探索

为了探索深度注意力模型在地铁出行预测任务中的可解释性,提出基于出行模式的注意力权重擦除方法和可解释性评估框架。利用提出的地铁出行深度注意力框架搭建预测模型,使用广州地铁羊城通数据构造三种不同长度出行序列数据集进行模型训练和验证,达到70%以上准确率;通过单一出行模式的注意力权重擦除实验发现,擦除最大注意力权重的出行模式比随机模式更能显著地影响模型预测结果,但大多数样本不发生预测结果的变化。即注意力机制在该条件下提供的可解释性信息是有限的,且该信息量随着序列长度增加而减小;通过一组出行模式注意力权重擦除实验结果表明,按注意力权重降序擦除能最快使模型预测结果发生变化,并且模型能稳定地对重要的出行模式的出行记录分配注意力权重,即注意力机制在该条件下较好地提供了可解释性信息,且该信息量随着序列长度增加而增大。     

空间桁架结构动力学形状优化设计

在多阶固有频率约束条件下,采用“渐进结点移动法”,对空间桁架结构的形状进行优化设计,使结构重量达到最小。首先分析固有频率相对结点位置的一阶导数,确定结点移动的效率,即灵敏度数。根据灵敏度分析结果,优先移动效率较高的结点。然后,利用库恩-塔克优化条件检验所得结果,保证优化过程收敛于最小重量设计。最后,用二个典型数值算例验证本文算法的有效性和可靠性。     

FTSR产线快冷模式下薄规格高强钢扁卷原因分析及控制

配置短距离、快速冷却层流系统的FTSR生产线具备较好的控轧控冷条件,可充分利用细晶强化生产高强度低合金热轧带钢,但薄规格钢卷普遍出现扁卷缺陷,影响后续加工6为探究扁卷机理,计算了试验钢的连续冷却转变曲线,并通过热模拟试验,研究了试验带钢轧后冷却的相变行为.结果表明,短距离快冷模式下热轧带钢在层冷线上相变孕育期缩短,相变...     

The influence of mining factors on seismic activity during longwall mining of a coal seam

In this article an attempt to determine the influence of mining factors on the seismic activity during the longwall mining of the upper layer of coal seam no. 405/2 in one of the Polish hard coal mines in the Upper Silesian Coal Basin was conducted. Two longwall panels were mined in analogous geological conditions and based on the same mining system and technology. However, there was significant difference with regards to the mining factors, which was reflected in the observed seismic activity. Some tools used in mining seismology were applied to illustrate the aforementioned influence of mining factors, e.g. the frequency-energy distribution, the frequency-magnitude distribution, the 2 D distribution of released seismic energy, the relationship between released seismic energy and the volume of mined coal, the Benioff strain release, and the Gutenberg-Richter(GR) b coefficient distribution(b is the proportion between high and low energy tremors). Concerning the Benioff strain release, a new solution, based on the slope of a fitted line in a moving time window, is proposed.     

One Mouse per Child: interpersonal computer for individual arithmetic practice

Single Display Groupware (SDG) allows multiple people in the same physical space to interact simultaneously over a single communal display through individual input devices that work on the same machine. The aim of this paper is to show how SDG can be used to improve the way resources are used in schools, allowing students to work simultaneously on individual problems at a shared display, and achieve personalized learning with individual feedback within different cultural contexts. We used computational fluency to apply our concept of ‘One Mouse per Child’. It consists of a participatory approach that makes use of personal feedback on an interpersonal computer for the whole classroom. This allows for N simultaneous intelligent tutoring systems, where each child advances at his or her own pace, both within a lecture and throughout the curricular units. Each student must solve a series of mathematical exercises, generated according to his or her performance through a set of pedagogical rules incorporated into the system. In this process, the teacher has an active mediating role, intervening when students require attention. Two exploratory studies were performed. The first study was a multicultural experience between two such distanced socio‐economic realities as Chile and India. It showed us that even in different environmental conditions, it is possible to implement this technology with minimal equipment (i.e. a computer, a projector, and one mouse per child). The second study was carried out in a third grade class in a low‐income school in Santiago de Chile. The students were asked to solve mainly addition exercises. We established statistically relevant results and observed that the software proved most beneficial for the students with the lowest initial results. This happens because the system adapts to the students' needs, reinforcing the content they most need to work on, thus generating a personalized learning process.     

Multi-level backcalculation algorithm for robust determination of pavement layers parameters

The backcalculation procedure applied to a mechanical characterization of the road pavement is usually limited to an identification of the elastic modulus of each layer only. The remaining parameters of the model are usually set as known, while performing an inverse analysis. Among assumed parameters are thicknesses of the model layers and frequently considered as constants on a homogeneous section of the road. This is an obvious simplification, because sections in general are inhomogeneous, i.e. the thickness of each layer changes slightly along each road section. Thus the precise and possibly nondestructive estimation of the layers thicknesses is very important and crucial in the inverse procedure. Here, a hybrid form of the optimization algorithm, where the condition of a constant thickness does not need to be fulfilled is described. Further on, the objective function is formed as a discrepancy between reference and computed deflection derivative instead of a deflection curve. In consequence, the values of backcalculated parameters are several per cent more precise compared to a standard procedure. Whenever the thickness of the asphalt layer of the pavement structure cannot be assumed a priori as a constant, the proposed here method appears to be necessary if one does not want to perform costly and destructive in situ drilling tests.     

基于VBA的公路用地自动化分类统计

针对公路用地统计繁琐的特点，提出了相关的处理方法。采用AutoCAD VBA编制了处理软件自动生成用地类别及面积。经某项目实际应用证明，该方法具有易于掌握、结果可靠、高效快捷的特点，可推广应用。     

Univariate and default standard unit biases in estimation of body weight and caloric content.

College students estimated the weight of adult women from either photographs or a live presentation by a set of models and estimated the calories in 1 of 2 actual meals. The 2 meals had the same items, but 1 had larger portion sizes than the other. The results suggest: (a) Judgments are biased toward transforming the example in question to the size and/or properties of a “standard” unit. For estimates of body weight, students assigned weights assuming a standard height, even though height information was provided in the photographs or directly present with live models. (b) There is an inclination to focus on 1 aspect or dimension of the stimulus (e.g., for female figures, their width, for meals the identity of the components as opposed to their size) and either devalue or completely ignore another parameter critical for accurate judgment (height, for the case of body weight estimations). That is, students defaulted to a normative unit size and thus treated the stimulus as a representative, categorical, and unvarying example, and focused on only 1 dimension (univariate bias) in making judgments. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Stochastic direct integration schemes for dynamic systems subjected to random excitations

In this work, a stochastic version of direct integration schemes is constructed, based on a general recursive state space formulation. The technique is applicable to evaluating the second order response statistics of systems subjected to non-stationary random excitations, and is potentially able to handle non-proportional damping. A specific group of these algorithms obtained using one of the most frequently used families of time stepping schemes is introduced. It is shown that statistics obtained for two multi-degree of freedom (MDOF) dynamic systems using the proposed method are in close agreement with those obtained from Monte Carlo simulation when a proper time step is used. Stability of the proposed scheme is systematically investigated and it is concluded that, for the response statistics to be bounded, the time step used in the algorithm has to lie within certain limits.