Women Awareness on the Importance of Agriculture Technologies Using ICT-Based Training Programs

Awareness policy intended to contribute to changing rural women realities to urgent needs of information and gain knowledge was to be demonstrated through in-depth information and communication technology-based(ICTs-based)training program that focused on the importance of advanced agricultural technologies in the production chain in developing countries like Egypt through access and use of the ICTs.Women are becoming well trained on the detailed steps of improved technologies applied in supply chain.Their increased awareness of the necessity of quality management to be followed during their work in the postharvest handling system helped them to produce high-quality products to meet the export requirements of foreign markets and add value to the export quality.Women have been able to reduce the extremely high losses that occurred due to improper handling in particular.The outcomes of proper and healthy procedures,precautions and personal protection were gained by rural women and technicians working in the supply chain.Moreover,women themselves became more confident in their know-how and more comfortable in transgressing cultural norms that inhibited their progress.

Cambricon-QR:a sparse and bitwise reproducible quantized training accelerator

Quantized training has been proven to be a prominent method to achieve deep neural network training under limited computational resources.It uses low bit-width arithmetics with a proper scaling factor to achieve negligible accuracy loss.Cambricon-Q is the ASIC design proposed to efficiently support quantized training,and achieves significant performance improvement.However,there are still two caveats in the design.First,Cambricon-Q with different hardware specifications may lead to different numerical errors,resulting in non-reproducible behaviors which may become a major concern in critical applications.Second,Cambricon-Q cannot leverage data sparsity,where considerable cycles could still be squeezed out.To address the caveats,the acceleration core of Cambricon-Q is redesigned to support fine-grained irregular data processing.The new design not only enables acceleration on sparse data,but also enables performing local dynamic quantization by contiguous value ranges(which is hardware independent),instead of contiguous addresses(which is dependent on hardware factors).Experimental results show that the accuracy loss of the method still keeps negligible,and the accelerator achieves 1.61×performance improvement over Cambricon-Q,with about 10%energy increase.

基于Tri-training的多特征融合图像检索

为了有效地综合利用图像的多种底层特征进行图像检索,提出将Tri-training方法应用于图像检索过程,将图像的颜色、纹理和形状特征进行了有效的融合。分别提取图像的三维量化颜色直方图、方向可控金字塔二值图像投影和仿射不变区域来表示其颜色、纹理和形状特征,并将三种特征的匹配值作为Tri-training分类器的输入对分类器进行训练和测试。实验结果表明,该方法有效利用了图像的多种特征,达到了很好的检索效果。

Critically Evaluate Foreign Language Teachers Training in Culture and Intercultural Communication

The aim of foreign language teaching in the contemporary is to develop foreign language talents who own intercultural competence.Thus,it is necessary to improve teachers’cultural quality and intercultural awareness.The paper explores some relevant training courses on culture and intercultural awareness,and discusses the deficiency of these courses.

Towards high performance low bitwidth training for deep neural networks

The high performance of the state-of-the-art deep neural networks(DNNs)is acquired at the cost of huge consumption of computing resources.Quantization of networks is recently recognized as a promising solution to solve the problem and significantly reduce the resource usage.However,the previous quantization works have mostly focused on the DNN inference,and there were very few works to address on the challenges of DNN training.In this paper,we leverage dynamic fixed-point(DFP)quantization algorithm and stochastic rounding(SR)strategy to develop a fully quantized 8-bit neural networks targeting low bitwidth training.The experiments show that,in comparison to the full-precision networks,the accuracy drop of our quantized convolutional neural networks(CNNs)can be less than 2%,even when applied to deep models evaluated on Image-Net dataset.Additionally,our 8-bit GNMT translation network can achieve almost identical BLEU to full-precision network.We further implement a prototype on FPGA and the synthesis shows that the low bitwidth training scheme can reduce the resource usage significantly.

Metacognitive Strategy Training in Reading Comprehension

Despite that language learning strategy(LLS) has been heatedly discussed since its emergence in 1975, its conceptualization, categorization and strategy training models are still in dispute. However, what commonly revealed by most studies is that there is a positive correlation between the use of language learning strategy and language learners' performance. This paper applies Dornyei's classification, known as cognitive strategies, metacognitive strategies, social strategies and affective strategies. By reviewing previous studies on the application of metacognitive strategy in English as a second language(ESL) reading comprehension, plenty of useful implications for reading pedagogy are suggested. It further explains how certain important metacognitive strategies, namely Semantic Mapping(SM) and Experience-Text-Relationship(ETR), can be implemented in second language reading comprehension, aiming to encourage language learners to be aware of learning strategies,"learn how to learn"(Cohen, 1988, p66)!

Concerns, performance, and awareness of people when experiencing haze and dust storms in Kermanshah

This study investigated the concerns and awareness of people in Kermanshah when haze and dust storms occur and also discussed the people’s performance during occurring and diffusion of this phenomenon. This descriptive-analytic study was done by component questionnaire and its sustainability and validity have been verified. The sample size of the statistical population of Kermanshah was selected as 400 people while it was considered that the population of the city was approximately 1 million people. The results showed that 77% of the participants in the study suffered from respiratory problems when the haze occurred. A total of 85.5% of the respondents in the present study believed that the precipitation rate had a direct effect to reduce haze. Also, 85% of the people with heart disorder had health problems during the haze. 63.3% of all participants had to use medical services when the haze occurred. However, 77.3% of participants were unaware of the necessary actions of the authorities when the haze occurred. Most of the studied people acknowledged that they obtained information about the negative effects of dust storms and the methods of reducing the effects to their health through broadcasting.

Seafarers＇ Current Awareness, Knowledge, Motivation and Ideas towards Low Carbon-Energy Efficient Operations

International and national concern about detrimental climate change has generated pressure for the shipping industry to play its＇ role in reducing the 3.3% of global carbon emission it emits. On the 1st January, 2013, the IMO （International Maritime Organisation） enforced regulations to support the reduction of shipping carbon emissions by improving energy efficiency. These measures directly and indirectly affect the daily operations of seafarers and onshore performance and managerial personnel. Whilst the industry has made efforts to raise the awareness of many stakeholders and research has been undertaken to investigate energy efficiency barriers, little has been done to capture the opinions, needs and knowledge of seafarers. A questionnaire was distributed in the last quarter of 2011 to investigate seafarers＇ awareness, knowledge and motivation towards carbon emissions in general and towards shipping carbon emissions. It also investigated opinions as to which personnel have the most influence over carbon reductions and what are the most important operational improvements that can be made. The authors have collected 317 questionnaire responses. The primary benefit of this study is to support the identification of an operational strategy to improve energy efficiency, including the development of LC-EE （low carbon-energy efficiency） MET （maritime education and training）, which is shown to be needed.

木材变磁铁创新趣味科研技能训练的探索与实践

高校是培养有见识、独立思考能力和创新意识专业人才的摇篮。通过木材变磁铁的科研技能训练课程实验,提升了学生综合运用所学知识开展科学研究的能力。本实验内容不但使学生了解木材的相关知识,而且锻炼了学生独立操作科研仪器和分析谱图的能力,打开了学生在科研领域的新视野,激发了学生热爱学习和科学研究的内在动力。

基于舍入误差的神经网络量化方法

深度神经网络需要付出高昂的计算成本,降低神经网络推理的功耗和延迟,是将神经网络集成到对功耗和计算严格要求的边缘设备上的关键所在。针对这一点,提出一种采用舍入误差的端到端神经网络训练后量化方法,缓解神经网络量化到低比特宽时带来的精度下降问题。该方法只需采用小批量且无标注的数据进行训练,且在不同的神经网络结构上都有十分不错的表现,RegNetX-3.2GF在权重和激活数的比特宽均为4的情况下分类准确率下降不到2%。

深度神经网络在线训练硬件加速器的数据量化综述

随着算法和数据的爆炸式增长,深度神经网络(Deep Neural Network, DNN)逐渐在实际应用中扮演愈发重要的角色。然而,真实场景中的数据与线下训练数据之间往往并不满足独立同分布假设,导致预训练DNN模型在实际应用中性能严重下降。所以,在资源供给相对有限的平台上进行DNN模型在线训练成为其有效应用的保证。为了满足真实场景对DNN模型质量与速度的多维度性能要求,如何在保证算法精度的同时显著降低计算复杂度是在此类应用中部署DNN的关键。数据量化是降低计算复杂度的主流优化技术之一,能够通过降低模型参数、中间值等数据的位宽来减少硬件加速器的资源耗费。因此,从软件和硬件两个方面对深度神经网络训练加速器中关于数据量化的研究进行总结。对国内外最新发表的相关文献进行归纳总结。首先,从软件的角度总结了不同的量化方法,包括简单映射数据量化和复杂映射数据量化;其次,从硬件的角度总结了DNN加速器对网络在线训练各计算步骤的量化支持;再次,阐述了数据量化对加速器设计的影响,包括存储单元和计算单元;最后,对本领域的研究进行总结,并展望了未来本领域的发展方向。文章提出的分类方法有助于对之前的DNN加速器在数据量化方面的工作进行分类。

基于注意力与量化感知的航拍红外目标检测

针对航拍场景下红外目标对比度低、识别精度差、检测难度大等问题,提出一种基于注意力与量化感知的航拍红外目标检测算法.首先,利用DCNv2替代ELAN模块中的3×3卷积,构建了DC-ELAN模块,有效提升了模型捕捉局部和全局特征的能力,进而强化了网络的特征表达能力;其次,通过巧妙地将SE注意力机制融入SPPCSPC模块和ELAN模块中,设计出了SE-SPPCSPC模块和SE-ELAN模块,有助于增强特征图的空间自注意力,模型能够更好地关注目标区域;此外,引入QARepVGG模块,提升模型的量化感知能力并增强其对量化误差的鲁棒性;最后,引入DyHead模块,该模块可以根据输入图像的不同动态调整检测头,提高模型对不同大小、形状目标的检测能力,从而进一步提高红外目标检测的准确性和鲁棒性.实验结果表明,相较于原模型,改进后的YOLOv7-tiny模型在计算量未增长的情况下, mAP@0.5值提升了3.4%, mAP@0.5:0.95值提升了4.8%,显著提高了模型检测精度.

一种基于自适应PoT量化的无乘法神经网络训练方法

当前的深度神经网络的训练过程中需要包含大量的全精度乘累加(MAC)操作,导致神经网络模型的线性层(包含卷积层和全连接层)的计算过程所需的能耗占整体能耗的绝大部分,达90%以上。本文提出了一种自适应逐层缩放的量化训练方法,可支持在神经网络计算全流程(前向传播和后向传播)将全部线性层中的全精度乘法替换为4位定点数加法计算和1位异或运算。实验结果表明,上述方法在能耗和准确率方面都优于现有方法,可支撑在训练过程中减少达95.8%的线性层能耗,在ImageNet数据集上的卷积神经网络和在WMT En-De任务上的Transformer网络得到小于1%的精度损失。

基于藏文字符感知的文本预训练模型方法研究

目前藏文预训练模型主要使用音节作为藏文单词表示。采用音节嵌入构建藏文单词表示,会存在藏文单词表示不完整且鲁棒性不高的问题。为了应对这一挑战,提出了一个名为藏文字符感知的预训练模型,该模型融合藏文字符、字丁和音节三个维度的特征,从藏文更细粒度的信息表征藏文单词特征。利用原始数据集和对抗性拼写错误测试集,评估了所提出的方法在藏文自动分词和命名实体识别任务上的性能。实验结果表明,该方法可以同时提高藏文预训练语言模型的性能和鲁棒性。

引入预训练表示混合矢量量化和CTC的语音转换

预训练模型通过自监督学习表示在非平行语料语音转换(VC)取得了重大突破。随着自监督预训练表示(SSPR)的广泛使用,预训练模型提取的特征中被证实包含更多的内容信息。提出一种基于SSPR同时结合矢量量化(VQ)和联结时序分类(CTC)的VC模型。将预训练模型提取的SSPR作为端到端模型的输入,用于提高单次语音转换质量。如何有效地解耦内容表示和说话人表示成为语音转换中的关键问题。使用SSPR作为初步的内容信息,采用VQ从语音中解耦内容和说话人表示。然而,仅使用VQ只能将内容信息离散化,很难将纯粹的内容表示从语音中分离出来,为了进一步消除内容信息中说话人的不变信息,提出CTC损失指导内容编码器。CTC不仅作为辅助网络加快模型收敛,同时其额外的文本监督可以与VQ联合优化,实现性能互补,学习纯内容表示。说话人表示采用风格嵌入学习,2种表示作为系统的输入进行语音转换。在开源的CMU数据集和VCTK语料库对所提的方法进行评估,实验结果表明,该方法在客观上的梅尔倒谱失真(MCD)达到8.896 d B,在主观上的语音自然度平均意见分数(MOS)和说话人相似度MOS分别为3.29和3.22,均优于基线模型,此方法在语音转换的质量和说话人相似度上能够获得最佳性能。

基于理论-实验-项目三位一体的制造工程认知实践课程改革

为了使学生在学习制造工程认识实践课程中,将理论知识的学习、实验技能的训练及实践能力的培养有机融合,提出构建理论-实验-项目三位一体的课题体系。理论部分主要起到扩大学生的知识面,了解制造业的前沿的目的;实验部分主要通过设计不同的实验环节,让学生掌握基本的实验技能;实践项目是基于学习的理论知识及掌握的实验技能,通过完整的项目选题、项目实施及项目展示培养学生的创新能力及实践动手能力。通过理论-实验-项目并行推进、融汇贯通,从而达到学生知识、技能、能力的全方位提升。同时建立与前期课程CAD和工程制图的衔接,打通由“工程表达”到“工程实现”的渠道,通过在课程中引入“精度”“配合”“成本”“效率”等概念,为学生后续系统学习机械设计、机械制造基础、先进制造实践等专业课程起到引导作用,从而实现课程的衔接与贯通。

试论高等师范院校“海洋地理学”教材建设

在我国实施海洋强国战略的时代背景下,高等师范院校的“海洋地理学”课程肩负着向地理师范生传授海洋地理知识、塑造正确海洋观的重任。然而,我国高等师范院校的“海洋地理学”课程教材建设明显滞后于快速发展的海洋地理学研究。从培养质量提升、海洋意识教育和学科体系需求三个角度探讨了高等师范院校“海洋地理学”课程教材建设的意义,同时给出了人海协调观和能力培养的教材建设理念,以期促进当前我国高等师范院校“海洋地理学”课程的教材建设。

基于时序信息的轻量级视频车辆目标检测方法

为实现在低功耗嵌入式设备上部署视频车辆目标检测模型,提出一种基于时序信息的轻量级视频车辆目标检测方法。该方法以SSD网络为基础,使用MobileNetV3-Small替换原主干特征提取网络VGG-16,并在SSD网络中直接注入注意力机制卷积GRU用于融合时序信息,提升车辆检测精度;关键帧检测网络控制的跳跃连接使模型只在关键帧更新GRU状态,非关键帧直接复制上一关键帧GRU状态,提升模型检测速度。为进一步减少计算量,网络中大量使用深度可分离卷积替换标准卷积层,同时使用量化感知训练方法压缩模型。在UA-DETRAC数据集上的实验表明,该方法在Intel Core i7 CPU和树莓派4B上平均每帧检测时间分别为18 ms和134 ms,准确率达到了较高水平,为78.81%。

在生物学教学中训练学生提问

在教学过程中,很多教师常常发现学生提出问题不够主动。因此,他们往往会自行提出相关问题并进行假设。然而,这表明教师的思维方式并未真正开放,从而无法有效地引导学生。本文探讨了在生物学教学过程中引导学生提出问题的思维训练策略,并提供了一些思考和建议。

基于量化的深度神经网络优化研究综述

随着模型参数和计算资源需求的不断增长,将模型部署在资源有限的设备上成为一个巨大的挑战。为解决这一挑战,量化成为了一种主要的方法,通过减少深度神经网络模型参数和中间过程特征图的位宽,可以对深度神经网络进行压缩和加速。文章全面回顾了基于量化的深度神经网络优化的工作原理。首先,讨论了常见的量化方法及其研究进展,并分析了各种量化方法之间的相似性、差异性以及各自的优缺点。其次,进一步探讨了分层量化、分组量化和通道量化等不同的量化粒度。最后,分析了训练与量化之间的相互关系,并讨论了当前研究所取得的成果和面临的挑战,旨在为未来深度神经网络量化研究提供理论基础。