基于改进MobileFaceNet的人脸识别方法北大核心

针对目前大多数人脸识别算法参数多、计算量大,难以部署到移动端和嵌入式设备中的问题,提出了一种基于改进MobileFaceNet的人脸识别方法。通过对MobileFaceNet模型结构的调整,将bottleneck模块优化为sandglass模块,改良深度卷积和逐点卷积的相对位置,适当增大sandglass模块的输出通道数,从而减少特征压缩时的信息丢失,增强人脸空间特征的提取。实验结果表明:改进后的方法在LFW测试数据集上准确率达99.15%,模型大小和计算量分别仅为原算法的61%和45%,验证了所提方法的有效性。     

Probabilistic learning on manifolds (PLoM) with partition

The probabilistic learning on manifolds (PLoM) introduced in 2016 has solved difficult supervised problems for the “small data” limit where the number N of points in the training set is small. Many extensions have since been proposed, making it possible to deal with increasingly complex cases. However, the performance limit has been observed and explained for applications for which N is very small and for which the dimension of the diffusion-map basis is close to N. For these cases, we propose a novel extension based on the introduction of a partition in independent random vectors. We take advantage of this development to present improvements of the PLoM such as a simplified algorithm for constructing the diffusion-map basis and a new mathematical result for quantifying the concentration of the probability measure in terms of a probability upper bound. The analysis of the efficiency of this extension is presented through two applications.     

机器翻译译文质量估计综述

机器翻译译文质量估计(Quality Estimation,QE)是指在不需要人工参考译文的条件下,估计机器翻译系统产生的译文的质量,对机器翻译研究和应用具有很重要的价值。机器翻译译文质量估计经过最近几年的发展,取得了丰富的研究成果。该文首先介绍了机器翻译译文质量估计的背景与意义;然后详细介绍了句子级QE、单词级QE、文档级QE的具体任务目标、评价指标等内容,进一步概括了QE方法发展的三个阶段: 基于特征工程和机器学习的QE方法阶段,基于深度学习的QE方法阶段,融入预训练模型的QE方法阶段,并介绍了每一阶段中的代表性研究工作;最后分析了目前的研究现状及不足,并对未来QE方法的研究及发展方向进行了展望。     

基于卷积神经网络的5G蜂窝网络无线定位方法

5G蜂窝网络发展迅猛，其覆盖面积将逐渐增大，因此使用5G蜂窝网络进行定位是有研究潜力的研究方向。本文提出一种新的深度学习技术来实现高效、高精度和低占用的定位，以代替传统指纹定位过程中繁重的指纹库生成以及距离计算。该方法建立了一个特殊的卷积神经网络，并根据5G天线信号的接收信号强度指示、相位和到达角等特征量，选择合适的输入数据格式构造样本组建训练集，对该卷积神经网络进行训练。训练得到的卷积神经网络可以替代指纹定位中的庞大指纹库，非常有利于直接在5G移动设备端实现定位。虽然卷积神经网络在训练过程中需要大量时间，但在训练完毕后直接进行分类定位的速度非常快，可以保障定位实现的实时性。本文所实现的卷积神经网络权重与偏置所占内存不到0.5 MB，且能够在实际应用环境中以95%的定位准确率以及0.1 m的平均定位精度实现高精度定位。     

快速近似计算Shapley值的归因解释方法

Shapley值归因解释方法虽然能更准确量化解释结果, 但过高的计算复杂度严重影响了该方法的实用性. 本文引入KD树重新整理待解释模型的预测数据, 通过在KD树上插入虚节点, 使之满足TreeSHAP算法的使用条件, 在此基础上提出了KDSHAP方法. 该方法解除了TreeSHAP算法仅能解释树结构模型的限制, 将该算法计算Shapley值的高效性放宽到对所有的黑盒模型的解释中, 同时保证了计算准确度. 通过实验对比分析, KDSHAP方法的可靠性, 以及在解释高维输入模型时的适用性.     

Neural Style Transfer for image within images and conditional GANs for destylization

In this paper, the feature representation of an image by CNN is used to hide the secret image into the cover image. The style of the cover image hides the content of the secret image and produce a stego image using Neural Style Transfer (NST) algorithm, which resembles the cover image and also contains the semantic content of secret image. The main technical contributions are to hide the content of the secret image in the in-between hidden layered style features of the cover image, which is the first of its kind in the present state-of-art-technique. Also, to recover the secret image from the stego image, destylization is done with the help of conditional generative adversarial networks (GANs) using Residual in Residual Dense Blocks (RRDBs). Further, stego images from different layer combinations of content and style features are obtained and evaluated. Evaluation is based on the visual similarity and quality loss between the cover-stego pair and the secret-reconstructed secret pair of images. From the experiments, it has been observed that the proposed algorithm has 43.95 dB Peak Signal-to-Noise Ratio (PSNR)), .995 Structural Similarity Index (SSIM), and .993 Visual Information Fidelity (VIF) for the ImageNet dataset. The proposed algorithm is found to be more robust against StegExpose than the traditional methods.     

Automated prediction system of household energy consumption in cities using web crawler and optimized artificial intelligence

The supply of electrical energy is critical to convenient and comfortable living. However, people consume a large amount of energy, contributing to an energy crisis and global warming, and damaging some ecological cycles. Residential electricity consumption has greater elasticity than industrial and business consumption; it therefore has high energy-saving potential. This work establishes an automated platform, which provides information about residential electricity consumption in each city in Taiwan. Machine learning was used to forecast future residential electricity demand. A nature-inspired optimization method was applied to enhance the accuracy of the best machine learner, yielding an even better hybrid ensemble model. Performance measures indicate that the resulting model is accurate and provides effective information for reference. An automatic web-based system based on the model was combined with a web crawler and scheduled to run automatically to provide information on monthly residential electricity consumption in each county and city. By providing energy consumption information across the country, power providers and government can discuss policy and set different goals for energy use. The results of this study can facilitate the early implementation of energy-saving and carbon emission-reducing in cities and aid utility companies in establishing energy conservation guidelines.     

Prediction of fish quality level with machine learning

In this study, sea bream, sea bass, anchovy and trout were captured and recorded using a digital camera during refrigerated storage for 7 days. In addition, their total viable counts (TVC) were determined on a daily basis. Based on the TVC, each fish was classified as ‘fresh’ when it was <5 log cfu per g, and as ‘not fresh’ when it was >7 log cfu per g. They were uploaded on a web-based machine learning software called Teachable Machine (TM), which was trained about the pupils and heads of the fish. In addition, images of each species from different angles were uploaded to the software in order to ensure the recognition of fish species by TM. The data of the study indicated that the TM was able to distinguish fish species with high accuracy rates and achieved over 86% success in estimating the freshness of the fish species tested.     

基于人工神经网络的智能化架空线路应变快速解调方法北大核心

为了提高智能化光纤复合架空线路态势感知的实时性,将人工神经网络方法应用于光纤沿线应变解调,确定了神经网络的结构。编程实现了基于洛伦兹模型的最小二乘谱拟合方法和神经网络方法,采用不同信噪比和布里渊频移的布里渊谱训练神经网络,将它们应用于某光纤复合架空线路沿线光纤应变的测量,从不同角度比较了两种方法的计算结果。计算结果表明,神经网络方法能有效获得光纤沿线的布里渊频移进而获得应变,具有与谱拟合方法相似的准确性,但应变解调时间仅约为谱拟合方法的1/20000。研究结果为提高智能光纤复合架空线路态势感知的实时性提供了参考。     

面向小语种机器翻译的平行语料库构建方法

神经机器翻译模型的训练效果在很大程度上取决于平行语料库的规模和质量。除了一些常见语言外,汉语与小语种间高质量平行语料库的建设一直处于滞后状态。现有小语种平行语料库多采用自动句子对齐技术利用网络资源构建而成,在文本质量和领域等方面有诸多局限性。采用人工翻译的方式可以构建高质量平行语料库,但是缺乏相关经验和方法。文中从机器翻译实践者和研究者角度出发,介绍了经济高效的人工构建小语种平行语料库的工作,包括其总体目标、实施过程、流程细节和最后结果。在构建过程中尝试并积累了各种经验,形成了小语种到汉语平行语料库构建方法、建议的总结。最终,成功构建了波斯语到汉语、印地语到汉语、印度尼西亚语到汉语各50万条高质量平行语料。实验结果表明,所构建的平行语料库有较好的质量,提高了小语种神经机器翻译模型的训练效果。