利用迁移学习的量化核最小平均P范数算法

在α稳定分布噪声背景下,核最小平均P范数算法(KLMP)的性能显著优于核最小均方算法(KLMS),但KLMP算法的计算量和存储容量都随迭代次数线性增加,不便实际应用。针对此问题,该文应用迁移学习理论,将基于样本实例获得的总滤波器划分为具有局部紧支撑结构的子滤波器之和,每个子滤波器的训练分别受不同的输入驱动,提出了最近实例质心估计核最小平均P范数算法(NICE-KLMP);为进一步减小存储容量,将在线矢量量化应用到该算法中,提出最近实例质心估计量化核最小平均P范数算法(NICE-QKLMP)。α稳定分布噪声背景下的 Mackey-Glass 时间序列预测的仿真结果表明,NICE-KLMP和NICE-QKLMP算法的复杂度显著低于KLMP算法,抗脉冲噪声性能显著强于NICE-KLMS算法,与KLMP算法相当。      

Semi-Supervised Learning Based on Manifold in BCI

Abstract-A Laplacian support vector machine （LapSVM） algorithm, a semi-supervised learning based on manifold, is introduced to brain-computer interface （BCI） to raise the classification precision and reduce the subjects＇ training complexity. The data are collected from three subjects in a three-task mental imagery experiment. LapSVM and transductive SVM （TSVM） are trained with a few labeled samples and a large number of unlabeled samples. The results confirm that LapSVM has a much better classification than TSVM.     

基于旋转学习机制的差分演化算法

为克服反向学习机制仅能搜索反向空间中一个固定点的弊端,通过引入旋转操作将其扩展为一种新的旋转学习机制,新机制通过调整旋转角度能搜索旋转空间中的任意一点,具备更强的勘探能力和多种应用模式.通过嵌入旋转学习算子,并引入参数自适应机制,提出了新的基于旋转学习的差分演化算法.在广泛使用的测试函数集上开展仿真实验,结果验证了旋转学习机制的有效性,与多种知名差分演化算法相比,新算法在寻优性能上竞争优势明显,且具有良好的适用性.     

基于迁移学习提高WiFi室内定位中信道状态信息指纹库的鲁棒性

基于信道状态信息(CSI)数据的WiFi指纹可用于室内定位。与信号强度值(RSSI)数据相比,CSI具有更高的数据信息粒度,并且可以在多个子载波上获得。当使用CSI数据进行室内定位时,相对于RSSI可以获得更好的结果。然而,无论使用RSSI还是CSI信号,在室内定位部署期间的一段时间后,室内环境通常会发生变化,并且基于测试数据的指纹数据库通常会恶化甚至失效。该文提出使用迁移学习算法来建立用于室内定位的指纹数据库。迁移学习的优势在于,可以使用较少的数据来获得更好的迁移训练结果。该文使用迁移学习来迁移指纹数据库的预测,延长指纹数据库的生命周期,并提高室内定位的鲁棒性。经过实验,1周后室内定位准确率保持在98%,两周后保持在97%。在相同成本下,该模型的生命周期和定位精度高于长短期记忆网络(LSTM)、卷积神经网络(CNN)、支持向量机(SVM)、深度神经网络(DNN)和其他定位系统。     

Transfer Learning Algorithm Design for Feature Transfer Problem in Motor Imagery Brain-computer Interface

The non-stationary of the motor imagery electroencephalography(MI-EEG) signal is one of the main limitations for the development of motor imagery brain-computer interfaces(MI-BCI). The nonstationary of the MI-EEG signal and the changes of the experimental environment make the feature distribution of the testing set and training set deviates,which reduces the classification accuracy of MI-BCI.In this paper, we propose a Kullback–Leibler divergence(KL)-based transfer learning algorithm to solve th...     

基于迁移集成学习的无人机图像识别算法

针对当前小型无人机目标图像识别方法准确率较低的问题,提出了一种基于迁移集成学习的无人机图像识别算法。首先,基于AlexNet、VGGNet-19、Inception-V3以及ResNet-50四种结构具有差异的卷积神经网络对源数据集进行预训练,获取图像的深层次特征;然后,对目标数据集进行迁移学习,得到目标的分类特征,构建分类模型;之后,采用相对多数投票法和加权平均法的集成学习方法,对分类模型进行集成得到迁移集成模型。构建了一个包含小型无人机图像、飞鸟图像以及直升机图像的图像数据集UavNet,在对数据集进行数据增强的基础上开展了图像识别算法性能实验,结果表明,算法对多类目标的识别准确率为99.42%,无人机类目标识别的F1-score指标为99.12%,优于主流的卷积神经网络方法和传统的支持向量机方法,具有一定的理论意义和应用价值。     

应用精英反向学习的多目标烟花爆炸算法

现实中的多目标优化问题越来越多,而且日益复杂.受混合多目标优化算法设计思想的启发,将烟花爆炸方法和精英反向学习机制引入至多目标优化领域,提出一种应用精英反向学习的多目标烟花爆炸算法(Multi-Objective Fireworks Optimization Algorithm Using Elite Opposition-Based Learning,MOFAEOL).该算法利用精英反向学习策略加强算法的全局搜索能力,利用烟花爆炸方法增强算法的局部搜索能力并提高求解的精度.这两种搜索机制相互协同以更好地平衡算法的全局勘探和局部开采的能力.MOFAEOL算法与另外5种代表性多目标优化算法一同在由ZDT系列和DTLZ系列组成的测试集上进行性能比较.实验表明,MOFAEOL算法在收敛性、多样性和稳定性方面均优于或部分优于其他对比算法.     

基于迁移演员-评论家学习的服务功能链部署算法

针对5G网络切片环境下由于业务请求的随机性和未知性导致的资源分配不合理从而引起的系统高时延问题,该文提出了一种基于迁移演员-评论家(A-C)学习的服务功能链(SFC)部署算法(TACA)。首先,该算法建立基于虚拟网络功能放置、计算资源、链路带宽资源和前传网络资源联合分配的端到端时延最小化模型,并将其转化为离散时间马尔可夫决策过程(MDP)。而后,在该MDP中采用A-C学习算法与环境进行不断交互动态调整SFC部署策略,优化端到端时延。进一步,为了实现并加速该A-C算法在其他相似目标任务中(如业务请求到达率普遍更高)的收敛过程,采用迁移A-C学习算法实现利用源任务学习的SFC部署知识快速寻找目标任务中的部署策略。仿真结果表明,该文所提算法能够减小且稳定SFC业务数据包的队列积压,优化系统端到端时延,并提高资源利用率。     

Automatic Image Annotation Using Semi-Supervised Multi-Instance Multi-Label Learning Algorithm

Automatic image annotation has emerged as an important research topic. From the perspective of machine learning, the annotation task fits both multiinstance and multi-label learning framework due to the fact that an image is composed of multiple regions, and is associated with multiple keywords as well. In this paper, we propose a novel Semi-supervised multi-instance multi-label （SSMIML） learning framework, which aims at taking full advantage of both labeled and unlabeled data to address the annotation problem. Specifically, a reinforced diverse density algorithm is applied firstly to select the Instance prototypes （IPs） with respect to a given keyword from both positive and unlabeled bags. Then, the selected IPs are modeled using the Gaussian mixture model （GMM） in order to reflect the semantic class density distribution. Furthermore, based on the class distribution for a keyword, both positive and unlabeled bags are redefined using a novel feature mapping strategy. Thus, each bag can be represented by one fixed-length feature vector so that the manifold-ranking algorithm can be used subsequently to propagate the corresponding label from positive bags to unlabeled bags directly. Experiments on the Corel data set show that the proposed method outperforms most existing image annotation algorithms.     

融合快照集成与知识迁移的人体行为识别算法

用卷积网络进行人体行为识别是毫米波雷达的一个热门研究方向。由于卷积网络结构的缺陷性,而且目前用于人体行为识别公开的雷达领域数据样本量过少,传统深度学习算法对雷达微多普勒图像的识别率不高,且在训练过程中容易出现过拟合的现象。针对上述问题,本文提出一种融合快照集成与迁移学习的雷达人体行为识别算法。首先,针对深度卷积网络无法提取图像全局特征的问题,该算法通过搭建Vision Transformer(VIT)模型引入注意力机制。其次,通过VIT模型在公开自然数据集上进行任务迁移和特征空间的迁移,解决微多普勒图像的识别过拟合的问题。最后,利用基于快照集成的投票机制算法,提升模型对复杂雷达微多普勒图像的识别能力。试验结果表明,在目标任务数据集样本量少、背景复杂的情况下,该算法能在不增加训练成本的前提下提升微多普勒图像的识别准确率,在VIT模型下该算法识别准确率达到了89.25%,优于经典卷积神经网络。     

Automatic Region-Based Image Annotation Using an Improved Multiple-Instance Learning Algorithm

Many existing image annotation algorithms work under probabilistic modeling mechanism. In this paper, we formulate the problem as a variation of supervised learning task and propose an Improved Citation kNN （ICKNN） Multiple-instance learning （MIL） algorithm for automatic image annotation. In contrast with the existing MIL based image annotation algorithm which intends to learn an explicit correspondence between image regions and keywords, here we annotate the keywords on the entire image instead of its regions. Concretely, we first explore the concept of Confidence weight （CW） for every training bag （image） to reflect the relevance extent between a bag and a semantic keyword. It can be treated as a stage of re-ranking on training set before an- notation starts. Moreover, a modified hausdoriT distance is adopted for the ICKNN algorithm to solve the automatic annotation problem. The proposed annotation approach demonstrates a promising performance over 5,000 images from COREL dataset, as compared with some current algorithms in the literature.     

基于最近邻聚类学习算法的模糊业务量预测器

本文根据模糊逻辑系统中的万能逼近定理,提出一个采用最近邻聚类学习长法的模糊业务量逼近器／预测器,通过对自回归（ＡＲ）过程及ＭＰＥＧ业务量的预测,了该逼近器／预测器的准确性和对业务量特性的适应性,最后讨论了该预测器在动态带宽分配中的应用。     

Classification of Breast Abnormalities Using a Deep Convolutional Neural Network and Transfer Learning

Journal of Communications Technology and Electronics - A new algorithm for classification of breast pathologies in digital mammography using a convolutional neural network and transfer learning is...     

Diagnostic Classification of Lung Cancer Using Deep Transfer Learning Technology and Multi-Omics Data

In recent years,with the increasing application of highthroughput sequencing technology,researchers have obtained and accumulated a large amount of multi-omics ...     

面向帕金森病语音诊断的非监督两步式卷积稀疏迁移学习算法

帕金森病(Parkinson's Disease,PD)语音诊断存在小样本问题,如果借助相关语音数据集进行迁移学习,容易加重训练集和测试集之间的分布差异,影响分类准确率.为了解决上述矛盾问题,本文提出了两步式稀疏迁移学习算法.该算法分为两大步:第一步算法为语音段特征同时优选的快速卷积稀疏编码算法,构造卷积稀疏编码算子用...     

An Algorithm for Idle-State Detection and Continuous Classifier Design in Motor-Imagery-Based BCI

Abstract-The development of asynchronous brain-computer interface （BCI） based on motor imagery （M1） poses the research in algorithms for detecting the nontask states （i.e., idle state） and the design of continuous classifiers that classify continuously incoming electroencephalogram （EEG） samples. An algorithm is proposed in this paper which integrates two two-class classifiers to detect idle state and utilizes a sliding window to achieve continuous outputs. The common spatial pattern （CSP） algorithm is used to extract features of EEG signals and the linear support vector machine （SVM） is utilized to serve as classifier. The algorithm is applied on dataset IVb of BCI competition Ⅲ, with a resulting mean square error of 0.66. The result indicates that the proposed algorithm is feasible in the first step of the development of asynchronous systems.     

A User-Friendly SSVEP-Based BCI Using Imperceptible Phase-Coded Flickers at 60Hz

A brain-computer interface(BCI) system based on steady-state visual evoked potentials(SSVEP) was developed by four-class phase-coded stimuli. SSVEPs elicited by flickers at 60 Hz, which is higher than the critical fusion frequency(CFF),were compared with those at 15 Hz and 30 Hz. SSVEP components in electroencephalogram(EEG) were detected using task related component analysis(TRCA)method. Offline analysis with 17 subjects indicated that the highest information transfer rate(ITR) was29.80±4.65 bp...     

An Algorithm for Idle-State Detection and Continuous Classifier Design in Motor-Imagery-Based BCI

The development of asynchronous braincomputer interface (BCI) based on motor imagery (MI) poses the research in algorithms for detecting the nontask states (i.e., idle state) and the design of continuous classifiers that classify continuously incoming electroencephalogram (EEG) samples. An algorithm is proposed in this paper which integrates two two-class classifiers to detect idle state and utilizes a sliding window to achieve continuous outputs. The common spatial pattern (CSP) algorithm is used to extract features of EEG signals and the linear support vector machine (SVM) is utilized to serve as classifier. The algorithm is applied on dataset IVb of BCI competition III, with a resulting mean square error of 0.66. The result indicates that the proposed algorithm is feasible in the first step of the development of asynchronous systems.