一个非全局连接神经网络模型

为了降低神经网络的连接复杂度,以利于硬件实现,本文提出了一个非全局连接的神经网络动力学模型,包括网络的能量函数、局域场和学习规则。基本思想是将Hopfield网络分解成若干个子网络,并建立各子网络之间的联系。对此模型的联想记忆过程统计分析表明,与Hopfield网络相比,一次分解后的连接数减少了25%,但网络的存储量和联想能力不变;多次分解后的连接数大大减少了,尽管网络的存储量和联想能力有所下降,但每个连接系数的平均存储能力提高了近1倍。本文最后讨论了高阶连接和多层网络结构的关系.

经典影视剧评论的计算传播分析——以爱奇艺《武林外传》为例

【目的】探究经典影视剧为何在新剧层出不穷的今天仍保持着较高热度,分析经典影视剧如何成为万千网友的情感纽带。【方法】借助Python编写程序收集《武林外传》剧集的评论,利用LDA主题模型、情感词典识别分析用户讨论的焦点和展现的情感,探究数字时代经典影视剧评论中的情感激荡。【结果】研究发现除了讨论剧情和角色,表达收获的快乐、满足,观众更多地通过讨论追忆个体经验,将过去、现在、未来相连接,展现出怀旧、陪伴等独特的情感。【结论】对普通观众而言,以《武林外传》为代表的经典剧集已超越剧集本身,融入万千网友的日常生活,成为万千网友和时代的独特见证。

基于连续语音识别技术的猪连续咳嗽声识别

针对现有基于孤立词识别技术的猪咳嗽声识别存在识别声音种类有限,无法反映实际患病猪连续咳嗽的问题,该文提出了基于双向长短时记忆网络-连接时序分类模型(birectional long short-termmemory-connectionist temporal classification,BLSTM-CTC)构建猪声音声学模型,进行猪场环境猪连续咳嗽声识别的方法,以此进行猪早期呼吸道疾病的预警和判断。研究了体质量为75 kg左右长白猪单个咳嗽声样本的持续时间长度和能量大小的时域特征,构建了声音样本持续时间在0.24～0.74 s和能量大于40.15 V^2·s的阈值范围。在此阈值范围内,利用单参数双门限端点检测算法对基于多窗谱的心理声学语音增强算法处理后的30 h猪场声音进行检测,得到222段试验语料。将猪场环境下的声音分为猪咳嗽声和非猪咳嗽声,并以此作为声学模型建模单元,进行语料的标注。提取26维梅尔频率倒谱系数(Mel frequency cepstral coefficients,MFCC)作为试验语段特征参数。通过BLSTM网络学习猪连续声音的变化规律,并利用CTC实现了端到端的猪连续声音识别系统。5折交叉验证试验平均猪咳嗽声识别率达到92.40%,误识别率为3.55%,总识别率达到93.77%。同时,以数据集外1 h语料进行了算法应用测试,得到猪咳嗽声识别率为94.23%,误识别率为9.09%,总识别率为93.24%。表明基于连续语音识别技术的BLSTM-CTC猪咳嗽声识别模型是稳定可靠的。该研究可为生猪健康养殖过程中猪连续咳嗽声的识别和疾病判断提参考。

Three dimensional finite element analysis of anatomic distal radius Nitinol memory connector treating distal radius fracture

Objective: To study the memory biomechanical character of anatomic distal radius Nitinol memory connector (DRMC) in treating distal radius fracture. Methods: Establishing three dimensional model and finite element analysis, we calculated the stress in and around the fracture faces when distal radius fracture was fixated with DRMC. Results: Axial holding stress produced by holding part of DRMC on distal radius was 14.66 MPa. The maximum stress of holding part was 40-70 MPa, the minimum stress was 3-7 MPa,and the stress of compression part was 20-40 MPa. Conclusion: The distribution of stress produced by DRMC around the fracture line is reasonable, and axial holding stress can help stabilize fracture during earlier period. The existence of longitudal compression and memory effect can transfer fixated disused section into developed section and enhance fracture healing.

基于LSTM-FCN的并发查询执行计划选择

查询是数据库系统的主要负载,为查询选择合适的执行计划是提高数据库系统性能、最终提升应用系统性能的关键。针对当前查询优化器为并发查询选择的执行计划准确率较低、动态性不足的问题,利用长短期记忆(longshort-termmemory,LSTM)网络的时域特性和全连接层网络(fullconnectednetworks,FCN)对特征的融合及分类优势,提出基于LSTM-FCN的并发查询执行计划选择方法。设计并编码查询组合的执行计划特征和交互特征,将其作为网络的输入,为查询动态选择适合实际运行场景的执行计划。在PostgreSQL上的实验验证了所提方法可行有效,LSTM-FCN在不同查询组合及并行度为3、4、5、6、7的情况下,以97.06%的平均准确率为查询选择合适的执行计划。

Research of Pipe Made of Cu -Zn -Si Shape Memory Alloy to Internally Sprayed Pipe Connection in Connected Technology

The copper-based shape memory alloy pipe joint was applied in the field of oil pipelines jointing as well as installing. It has many advantages than anything. In the paper, the basic principle and connection influence factors of copper-based shape memory alloy pipe joint were described. Experimental results shows the copper-based shape memory alloy pipe joints can be used in oilfield oil pipeline connection and it can solve the inner pipe joint corrosion damage problem by welding.