The anatomy of vampire

We present an implementation technique for a class of bottom-up logic procedures. The technique is based oncode trees. It is intended to speed up most important and costly operations, such as subsumption and resolution. As an example, we consider the forward subsumption problem which is the bottleneck of many systems implementing first-order logic.To efficiently implement subsumption, we specialize subsumption algorithms at run time, using theabstract subsumption machine. The abstract subsumption machine makes subsumption-check using sequences of instructions that are similar to the WAM instructions. It gives an efficient implementation of the clause at a time subsumption problem. To implement subsumption on the set at a time basis, we combine sequences of instructions incode trees.We show that this technique yields a new way of indexing clauses. Some experimental results are given.The code trees technique may be used in various procedures, including binary resolution, hyper-resolution, UR-resolution, the inverse method, paramodulation and rewriting, OLDT-resolution, SLD-AL-resolution, bottom-up evaluation of logic programs, and disjunctive logic programs.Supported by Swedish TFR grant No. 93–409     

自底向上构建高效BVH的研究

光线追踪中,加速结构对于减少光线与几何体求交的计算量起到了不可或缺的作用,找到一种高效的加速结构仍然是本领域的研究热点之一.传统的加速结构如BVH,都是自顶向下的构建方式,由于其本身的限制,构造出的加速结构往往不是最优的.为了改进自顶向下的缺陷,提出了一种自底向上BVH的构建方式,它使用了BVH和空间划分的混合结构,并引入了一种改进的表面积启发代价函数和多个构建参数来优化最后的构建结果.最后实验表明,此文中提出的方式比传统的BVH与KD-Trees有更好的加速效果,可以实现对前两者5%-10%的提速.最后,文中还给出了一些加速方案来减少构建时间.     

结合自底向上注意力机制和记忆网络的视觉问答模型

目的 现有大多数视觉问答模型均采用自上而下的视觉注意力机制,对图像内容无加权统一处理,无法更好地表征图像信息,且因为缺乏长期记忆模块,无法对信息进行长时间记忆存储,在推理答案过程中会造成有效信息丢失,从而预测出错误答案。为此,提出一种结合自底向上注意力机制和记忆网络的视觉问答模型,通过增强对图像内容的表示和记忆,提高视觉问答的准确率。方法 预训练一个目标检测模型提取图像中的目标和显著性区域作为图像特征,联合问题表示输入到记忆网络,记忆网络根据问题检索输入图像特征中的有用信息,并结合输入图像信息和问题表示进行多次迭代、更新,以生成最终的信息表示,最后融合记忆网络记忆的最终信息和问题表示,推测出正确答案。结果 在公开的大规模数据集VQA （visual question answering）v2.0上与现有主流算法进行比较实验和消融实验,结果表明,提出的模型在视觉问答任务中的准确率有显著提升,总体准确率为64.0%。与MCB（multimodal compact bilinear）算法相比,总体准确率提升了1.7%;与性能较好的VQA machine算法相比,总体准确率提升了1%,其中回答是/否、计数和其他类型问题的准确率分别提升了1.1%、3.4%和0.6%。整体性能优于其他对比算法,验证了提出算法的有效性。结论 本文提出的结合自底向上注意力机制和记忆网络的视觉问答模型,更符合人类的视觉注意力机制,并且在推理答案的过程中减少了信息丢失,有效提升了视觉问答的准确率。     

基于改进Mask R-CNN的建筑钢筋尺寸检测算法

建筑施工现场钢筋图像背景复杂且干扰较多,传统图像检测算法无法有效利用特征信息,难以满足现阶段建筑智能监理行业中钢筋尺寸检测精度的验收要求。提出一种在Mask R-CNN模型基础上加入自下而上路径和注意力机制的改进模型BU-CS Mask R-CNN。在建筑工地现场拍摄图像后,整理自建钢筋数据集,并在此数据集上进行算法验证。实验结果表明,与Mask R-CNN模型相比,BU-CS Mask R-CNN模型的召回率、交并比和像素准确率分别提升了4.9%、6.8%、7.4%,钢筋直径和间距的尺寸检测精度分别提升了14.9%、4.4%,能得到更加准确的钢筋目标检测框和边缘分割掩膜,达到了行业中实际工程验收的精度要求。     

邻近地铁隧道的软土深基坑变形实测分析

上海外滩596地块超深基坑紧邻地铁9号线区间隧道及一系列管线和建筑物。为控制基坑施工对周边环境(尤其地铁隧道)的影响,本项目设计采取分坑顺作、两墙合一地下连续墙、钢支撑轴力补偿体系、被动区加固、抽条分块开挖等系列措施。实测结果表明,远离地铁侧的地下连续墙最大变形为45.6 mm,邻近地铁侧地下连续墙最大变形为17.2 mm,邻近地铁隧道的最大隆起量为12.9 mm。所采用的设计方案满足了地铁的变形控制要求。     

Analysis of catchment evapotranspiration at different scales using bottom-up and top-down approaches

Physically-based hydrological models are used to predict catchment water balance through detailed simulation of hydrological processes at small temporal and spatial scales. However, annual catchment water balance can also be easily and simply predicted using lumped conceptual model. Comparison between physically-based hydrological models and lumped conceptual models can help us understand the dominant factors on catchment water balance at different scales. In this paper, a distributed physically-based hydrological model (i.e., bottom-up approach) and a simple water-energy balance model (i.e., top-down approach) are used to predict actual evapotranspiration in nine sub-catchments, and the whole basin of the Luan River in northern China. Both simulations give very close values of annual evapotranspiration and show the same complementary relationship between actual and potential evapotranspiration at annual time scale. From the analysis at different time scales through comparison of the top-down and the bottom-up methods, it is shown that the annual catchment evapotranspiration is controlled mainly by annual precipitation and potential evapotranspiration, and the variability of soil water and vegetation becomes more important at a smaller time scale in the study areas. It is also known that the relationship between potential and actual evapotranspiration shows a highly nonlinear relationship at the annual and catchment scale but can be simplified to a linear relationship at hourly temporal and hillslope scales, which is commonly used in the physically-based hydrological models.     

北京市“十三五”时期民用建筑节能潜力研究

运用"自下而上"节能措施法,通过新建建筑采暖节能、既有建筑节能改造、可再生能源建筑应用、公共建筑节能监管等节能措施,得到"十二五"外推、强化政策以及极端政策三种情景下,将分别产生节能潜力391.31,602.95和804.41万t标煤。采用"自上而下"能耗分析法,得到由人口总量和建筑面积总量控制能够产生的节能潜力236.67万t标煤,仅次于城镇新建建筑采暖节能潜力。因此,在"十三五"建筑节能工作中,不仅要重视建筑节能措施本身,还要考虑相关宏观因素的作用,从更大范围和更广视角来对待建筑节能工作。     

Modular synthesis of Petri nets for modeling flexible manufacturing systems

This paper proposes a modular Petri net synthesis method for modeling flexible manufacturing systems based on synchronization among control processes of the manufacturing resources (such as robots and machines). In the method, the target system is modeled in a bottom-up and uniform manner by first describing the system's control processes using strongly connected state machines (SCSMs) as the basic modules. Each SCSM may contain multiple tokens to represent resources from the same type such as spaces in a buffer. Next, the common transitions and common transition subnets of the modules are merged to represent their synchronization. The system model constructed is proven to be conservative and thus bounded. Moreover, a restricted class of merged nets is proven to be live and reversible. For general classes of merged nets, this paper shows theorems that easily calculateP-invariants of the final net without solving the linear system equations. TheseP-invariants can be used to help in verifying the model's qualitative properties such as liveness.     

Visual function and cognitive aging: Differential role of contrast sensitivity in verbal versus spatial tasks.

Cognitive and sensory function are correlated in older adults. Sensory function may provide an index of neurological integrity (common-cause hypothesis). Declining sensory input may also directly impair cognition (direct-cause hypothesis). Accordingly, sensory function should more strongly predict cognitive performance and should account for more age-related variability in tasks with higher sensory demands. In a cross-sectional adult life span sample, visual contrast sensitivity was a better predictor and accounted for more of the age-related variability in high sensory-demand tasks, compared with low sensory-demand tasks, consistent with the direct-cause hypothesis. The results suggest a direct role for sensory function in cognitive aging when task conditions place heavy demands on sensory processing. (PsycINFO Database Record (c) 2010 APA, all rights reserved)