基于深度学习的无人机巡检图像销钉故障检测

随着经济和社会的发展,发电量和用电量逐年上升。安全的电力保障关系到国计民生,在常年的使用过程中,由于电力传输的输电线路受到外界环境的影响,使得输电线路部件容易出现不同程度的破损,其中销钉是固定螺母的关键零件,销钉的脱落会导致各部件之间连接的不稳定,这给输电网络的安全运行带来了极大的隐患。随着深度学习技术在计算机视觉领域中的应用,使得机器自动识别销钉这一输电线路系统中的微小部件成为现实。采用Faster R-CNN算法对无人机巡检图像中的销钉脱落故障进行识别,并讨论了不同分类器对识别结果的影响,然后对ACF+Adaboost、Hough+LSD和Faster R-CNN检测方法进行比较。实验结果表明,基于Faster R-CNN的目标检测方法对于输电线路中销钉脱落故障的识别率可达到96%,同时对正常销钉的识别率最高可达98%。     

面向隐私保护联邦学习的医学影像目标检测算法

联邦学习技术是一种新型多机构协同训练模型范式,广泛应用于多领域,其中模型参数隐私保护是一个关键问题.针对CT影像综合性病灶检测任务,提出隐私保护的联邦学习算法.首先部署松散耦合的客户端-服务器架构;其次在各客户端使用改进的RetinaNet检测器,引入上下文卷积和后向注意力机制;最后完成联邦训练.各客户端使用局部更新策略,采用自适应训练周期,局部目标函数中加入了限制项;服务器使用自适应梯度裁剪策略和高斯噪声差分隐私算法更新全局模型参数.在DeepLesion数据集上的消融分析说明了算法各部分的重要性.实验结果表明,改进的RetinaNet检测器有效地提升了多尺度病灶的检测精度.与集中数据训练模型范式相比,联邦学习所得模型性能略低(mAP分别为75.33％和72.80％),但训练用时缩短近38％,有效地实现了隐私保护、通信效率和模型性能的良好权衡.     

基于纵向联邦学习的社交网络跨平台恶意用户检测方法

近年来,在线社交网络恶意用户呈现出分散性、潜伏性、复杂性等特征,如何在保障普通用户数据隐私的前提下,融合多方数据进行建模分析,实现对恶意用户的精确检测成为研究人员关注的焦点.本文提出了一种基于纵向联邦学习的社交网络跨平台恶意用户检测方案.首先,通过对多源异构数据进行预处理,采用加密样本对齐和加密模型训练方法,构建了基于纵向联邦学习的跨平台恶意用户检测层次化架构;其次,对安全联邦提升树算法进行分析和改进,提出了一种面向多方隐私保护的恶意用户检测算法;最后,基于现实社交网络平台实验研究分析,所提出的方案不仅具有安全性,而且模型算法相较于其他两个基线模型,准确率分别提升了14.03%和1.918%.     

基于GB-AEnet-FL网络的物联网设备异常检测

针对物联网场景下,传统异常检测方法在海量不均衡数据中检测准确率低、数据异构导致模型泛化能力差等问题,提出了基于联邦学习的对抗双编码异常检测网络 （GB-AEnet-FL）的物联网设备异常检测算法。首先,提出了一种基于异常数据的主动特征分布学习算法,主动学习数据的潜在特征分布,通过数据重构扩充异常数据,均衡正负样本比例。其次,在潜在特征层引入了对抗训练机制并添加一致性增强约束和收缩约束,提高特征提取的精度。最后,设计了一种基于动态模型选择的联邦学习算法,比较局部模型与全局模型的置信度评分,动态选择部分联邦体参与,加速模型的聚合,在一定程度上也保护了用户隐私。在四个不同数据集上进行验证,结果显示,所提算法在检测准确度优于传统算法,且泛化能力得到相应提升。     

面向非独立同分布数据的联邦梯度提升决策树

随着联邦学习的不断兴起,梯度提升决策树（GBDT）作为一种传统的机器学习方法,逐渐应用于联邦学习中以达到理想的分类效果。针对现有GBDT的横向联邦学习模型,存在精度受非独立同分布数据的影响较大、信息泄露和通信成本高等问题,提出了一种面向非独立同分布数据的联邦梯度提升决策树（federated GBDT for non-IID dataset,nFL-GBDT）。首先,采用局部敏感哈希（LSH）来计算各个参与方之间的相似样本,通过加权梯度来构建第一棵树。其次,由可靠第三方计算只需要一轮通信的全局叶权重来更新树模型。最后,实验分析表明了该算法能够实现对原始数据的隐私保护,并且通信成本低于simFL和FederBoost。同时,实验按照不平衡比率来划分三组公共的数据集,结果表明该算法与Individual、TFL及F-GBDT-G相比,准确率分别提升了3.53%、5.46%和4.43%。     

基于自监督学习的输电线路螺栓螺母异常检测技术

针对输电线路螺栓螺母异常检测问题,对无标签螺栓数据训练深度学习预训练模型进行了研究,首次采用了自监督学习的方法,使用大量无标签数据集进行学习,首先,使用少量带标签的螺栓目标检测数据集训练目标检测模型,推理获取大量的单图无标签螺栓数据集,然后,采用自监督学习方法通过单图无标签螺栓数据集训练螺栓预训练大模型并获取可视化的螺栓注意力图,最后通过实验对比螺栓预训练大模型在螺栓分类和检索任务中与非预训练模型的效果,实验结果表明无标签数据得到的预训练大模型可以注意到螺杆、螺母、连接件的位置,进一步表明螺栓预训练大模型在分类任务中准确率提升了2%到7%,在螺栓检索任务中平均精度提升了8%。     

基于联邦学习和深度残差网络的入侵检测

深度学习被广泛应用到入侵检测领域,但大多数研究的重点是通过改进算法提高入侵检测的准确率,却忽视了在实际应用中单个用户拥有的数据无法满足训练需求的问题。为了实现网络入侵检测模型在训练过程中保护用户隐私安全的同时,仍具有对网络流量数据检测异常的能力,提出一种基于联邦学习并融合深度残差网络（ResNet）和注意力机制的入侵检测模型FL-SEResNet(Federation Learning Squeeze-and-Excitation network ResNet)。在训练过程中,通过对数据压缩、解压、分发、加密和聚合等操作,可以在保护参与者数据隐私的同时,通过多方参与提供足够的训练数据。在NSL-KDD和UNSW-NB15数据集上,所提模型在多分类实验的识别准确率分别为84.22%和80.38%。在NSL-KDD上,与同属于联邦学习的CNN-FL相比,对多分类的识别准确率提升了1.82个百分点,对少数类R2L(Remote to Local)的识别准确率提升了24.94个百分点。     

基于数据分布的聚类联邦学习

联邦学习(federated learning)可以解决分布式机器学习中基于隐私保护的数据碎片化和数据隔离问题。在联邦学习系统中,各参与者节点合作训练模型,利用本地数据训练局部模型,并将训练好的局部模型上传到服务器节点进行聚合。在真实的应用环境中,各节点之间的数据分布往往具有很大差异,导致联邦学习模型精确度较低。为了解决非独立同分布数据对模型精确度的影响,利用不同节点之间数据分布的相似性,提出了一个聚类联邦学习框架。在Synthetic、CIFAR-10和FEMNIST标准数据集上进行了广泛实验。与其他联邦学习方法相比,基于数据分布的聚类联邦学习对模型的准确率有较大提升,且所需的计算量也更少。     

基于深度学习和部分模型的相关性人脸检测

针对人脸检测中的部分遮挡和多姿态问题,提出一种基于深度学习的人脸检测算法。首先利用部件检测器检测人脸局部区域,然后将人脸局部区域检测结果输入到深度模型中,学习各局部区域之间的相关性,完成人脸检测。该算法将深度学习理论与基于部分模型的思想相结合,实现人脸检测。实验结果表明,该算法在遮挡、多姿态等情况下具有良好的鲁棒性。     

融合注意力机制与联合优化的表面缺陷检测

两段式缺陷检测模型中分割和分类网络的优化目标不一致,导致二者耦合性较差,且分割模块误差的积累可能进一步弱化分类模块的性能.针对上述问题,提出一种基于注意力机制的缺陷检测联合优化算法.首先基于混合注意力特征融合模块的分割网络融合浅层特征和深层特征,提取更全面的缺陷位置信息;然后基于多感受野空间注意力模块的分类网络挖掘更具判别性的缺陷类别特征;最后通过联合优化目标实现分割和分类网络的学习优化,提升整个算法的耦合性以及性能.基于PyTorch框架,在公开工业缺陷检测数据集DAGM 2007, MAGNETIC-TILE和KolektorSDD2数据集上进行实验,并引入分段式算法及类U-Net算法进行横向对比的结果表明,所提算法的准确率相比分段式算法最高提升28.02%,相比类U-Net算法最高提升8.3%,且精确率、召回率、F1值均优于同类算法,具有更好的检测性能.     

On the Complexity of Function Learning

The majority of results in computational learning theory are concerned with concept learning, i.e. with the special case of function learning for classes of functions with range {0, 1}. Much less is known about the theory of learning functions with a larger range such as or . In particular relatively few results exist about the general structure of common models for function learning, and there are only very few nontrivial function classes for which positive learning results have been exhibited in any of these models.We introduce in this paper the notion of a binary branching adversary tree for function learning, which allows us to give a somewhat surprising equivalent characterization of the optimal learning cost for learning a class of real-valued functions (in terms of a max-min definition which does not involve any learning model).Another general structural result of this paper relates the cost for learning a union of function classes to the learning costs for the individual function classes.Furthermore, we exhibit an efficient learning algorithm for learning convex piecewise linear functions from ^d into . Previously, the class of linear functions from ^d into was the only class of functions with multidimensional domain that was known to be learnable within the rigorous framework of a formal model for online learning.Finally we give a sufficient condition for an arbitrary class of functions from into that allows us to learn the class of all functions that can be written as the pointwise maximum ofk functions from . This allows us to exhibit a number of further nontrivial classes of functions from into for which there exist efficient learning algorithms.     

The Power of Self-Directed Learning

This article studies self-directed learning, a variant of the on-line (or incremental) learning model in which the learner selects the presentation order for the instances. Alternatively, one can view this model as a variation of learning with membership queries in which the learner is only charged for membership queries for which it could not predict the outcome. We give tight bounds on the complexity of self-directed learning for the concept classes of monomials, monotone DNF formulas, and axis-parallel rectangles in {0, 1, , n – 1} ^d . These results demonstrate that the number of mistakes under self-directed learning can be surprisingly small. We then show that learning complexity in the model of self-directed learning is less than that of all other commonly studied on-line and query learning models. Next we explore the relationship between the complexity of self-directed learning and the Vapnik-Chervonenkis (VC-)dimension. We show that, in general, the VC-dimension and the self-directed learning complexity are incomparable. However, for some special cases, we show that the VC-dimension gives a lower bound for the self-directed learning complexity. Finally, we explore a relationship between Mitchell's version space algorithm and the existence of self-directed learning algorithms that make few mistakes.     

Learning from a Population of Hypotheses

We introduce a new formal model in which a learning algorithm must combine a collection of potentially poor but statistically independent hypothesis functions in order to approximate an unknown target function arbitrarily well. Our motivation includes the question of how to make optimal use of multiple independent runs of a mediocre learning algorithm, as well as settings in which the many hypotheses are obtained by a distributed population of identical learning agents.     

Indexing,Elaboration and Refinement: Incremental Learning of Explanatory Cases

This article describes how a reasoner can improve its understanding of an incompletely understood domain through the application of what it already knows to novel problems in that domain. Case-based reasoning is the process of using past experiences stored in the reasoner's memory to understand novel situations or solve novel problems. However, this process assumes that past experiences are well understood and provide good lessons to be used for future situations. This assumption is usually false when one is learning about a novel domain, since situations encountered previously in this domain might not have been understood completely. Furthermore, the reasoner may not even have a case that adequately deals with the new situation, or may not be able to access the case using existing indices. We present a theory of incremental learning based on the revision of previously existing case knowledge in response to experiences in such situations. The theory has been implemented in a case-based story understanding program that can (a) learn a new case in situations where no case already exists, (b) learn how to index the case in memory, and (c) incrementally refine its understanding of the case by using it to reason about new situations, thus evolving a better understanding of its domain through experience. This research complements work in case-based reasoning by providing mechanisms by which a case library can be automatically built for use by a case-based reasoning program.     

基于移动学习的学科主题学习资源设计研究

移动学习作为一种新型的学习方式正成为研究热点,而基于移动学习的学科主题学习资源相对缺乏。本文阐述了移动学习的概念及特点、主题学习、学科主题学习资源的理论基础,分析了基于移动学习的学科主题学习资源设计的基本原则,最后构建了基于移动学习的学科主题学习资源的设计框架。     

Exactly Learning Automata of Small Cover Time

We present algorithms for exactly learning unknown environments that can be described by deterministic finite automata. The learner performs a walk on the target automaton, where at each step it observes the output of the state it is at, and chooses a labeled edge to traverse to the next state. The learner has no means of a reset, and does not have access to a teacher that answers equivalence queries and gives the learner counterexamples to its hypotheses. We present two algorithms: The first is for the case in which the outputs observed by the learner are always correct, and the second is for the case in which the outputs might be corrupted by random noise. The running times of both algorithms are polynomial in the cover time of the underlying graph of the target automaton.     

学习控制技术·方法应用的发展新动向

分析和概述了当前学习控制系统所采用的技术、学习方法及应用的发展新动向 .从所采用的技术来看 ,学习控制正在从采用单一的技术向采用混合技术的方向发展 ;从学习方法和应用来看 ,学习控制正在从采用较为简单的参数学习向采用较为复杂的结构学习、环境学习和复杂对象学习的方向发展     

基于深度神经网络的少样本学习综述

如何从少数训练样本中学习并识别新的类别对于深度神经网络来说是一个具有挑战性的问题。针对如何解决少样本学习的问题,全面总结了现有的基于深度神经网络的少样本学习方法,涵盖了方法所用模型、数据集及评估结果等各个方面。具体地,针对基于深度神经网络的少样本学习方法,提出将其分为数据增强方法、迁移学习方法、度量学习方法和元学习方法四种类别;对于每个类别,进一步将其分为几个子类别,并且在每个类别与方法之间进行一系列比较,以显示各种方法的优劣和各自的特点。最后强调了现有方法的局限性,并指出了少样本学习研究领域未来的研究方向。     

The comparative effect of individually-constructed vs. collaboratively-constructed computer-based concept maps

The researchers investigated the comparative effects of individually-constructed and collaboratively-constructed computer-based concept mapping on middle school science concept learning. One hundred and sixty one students completed the entire study. Using prior science performance scores to assure equivalence of student achievement across groups, students were assigned to three groups: a self-selected study strategy group, an individual-concept mapping group, and a collaborative pairs – concept mapping group. Collaboratively and individually-constructing computer-based concept maps had equally positive effects on seventh grade middle school science concept learning as measured on a comprehension test. However, the students who collaboratively constructed concept maps created significantly higher quality concept maps than those who individually constructed concept maps indicating deeper conceptual understanding.     

Usage of a mobile social learning platform with virtual badges in a primary school

This study presents and examines SamEx, a mobile learning system used by 305 students in formal and informal learning in a primary school in Singapore. Students use SamEx in situ to capture media such as pictures, video clips and audio recordings, comment on them, and share them with their peers. In this paper we report on the experiences of students in using the application throughout a one-year period with a focus on self-directedness, quality of contributions, and answers to contextual question prompts. We examine how the usage of tools such as SamEx predicts students' science examination results, discuss the role of badges as an extrinsic motivational tool, and explore how individual and collaborative learning emerge. Our research shows that the quantity and quality of contributions provided by the students in SamEx predict the end-year assessment score. With respect to specific system features, contextual answers given by the students and the overall likes received by students are also correlated with the end-year assessment score.