In recent years,portfolio management problems have been extensively studied in the field of artificial intelligence,but there are some improvements in the existing quantitative trading methods based on deep learning.First of all,the prediction model of stocks is single,usually a model only trains a trading expert,and the decision of trading is only based on the prediction results of the model.Secondly,the data source used in the model is relatively single,only considering the stock's own data,ignoring the impact of the entire market risk on the stock.Aiming at the above problems,a reinforcement learning model based on dynamic se-lection predictor(DSDRL)is proposed.The model is divided into three parts.Firstly,the characteristics of stock data are extrac-ted and introduced into multiple predictors.Multiple prediction models are trained for different investment strategies,and the cur-rent optimal prediction results are obtained by dynamic selector.Secondly,the market environment evaluation module is used to quantify the current market risk and obtain the appropriate proportion of investment amount.Finally,based on the first two mo-dules,a deep reinforcement learning model is established to simulate the real trading environment,and the actual portfolio strate-gy is obtained based on the predicted results and the proportion of investment amount.In this paper,the daily k-line data of China Securities 500 and S & P 500 are used for test verification.The results show that the proposed model is superior to other refe-rence models in Sharpe rate and other indicators.
Time series anomaly detection is one of the critical tasks in time series analysis. There are many existing works using autoencoders for time series anomaly detection; however, autoencoders can not only reconstruct normal sequences but also reconstruct anomalous sequences sometimes, which leads to the poor effect of anomaly detection based on reconstruction error. At the same time, the real-world time series have the characteristic of temporal coupling, and the existing methods often cannot take into account the multiple modes of the time series, which leads to the low accuracy of time series anomaly detection. Aiming at the above problems, we propose a Time Series Anomaly Detection method incorporating Wavelet Decomposition and Temporal Decoupled Autoencoder (WDAE). Specifically, WDAE first reshapes the original 1D time series into 2D data harboring different temporal patterns based on frequency components and independently extracts the different temporal features of the time series data. Then, the original data are reconstructed in the time domain by the autoencoder to obtain the reconstructed time-domain sequence, and in the frequency domain, the 2D data are subjected to the discrete wavelet transform and the inverse wavelet transform after weight adjustment to obtain the reconstructed frequency domain sequence, and the captured normal sequence pattern in the frequency domain restricts the latent feature space in the time domain so that the autoencoder can rebuild the normal sequence to avoid reconstructing the anomalous sequence well. Experiments on five publicly available data show that WDAE significantly improves the F1 value over related state-of-the-art baseline methods.
In order to capture and integrate the structural features and temporal features contained in social graph and diffusion cascade more effectively, an information diffusion prediction model based on the Transformer and Relational Graph Convolutional Network (TRGCN) is proposed. Firstly, a dynamic heterogeneous graph composed of the social network graph and the diffusion cascade graph was constructed, and it was input into the Relational Graph Convolutional Network (RGCN) to extract the structural features of each node. Secondly, the time embedding of each node was reencoded using Bi-directional Long Short-Term Memory (Bi-LSTM). The time decay function was introduced to give different weights to nodes at different time positions, so as to obtain the temporal features of nodes. Finally, structural features and temporal features were input into Transformer and then merged. The spatiotemporal features are obtained for information diffusion prediction. The experimental results on three real datasets of X (formerly known as Twitter), Douban, and Memetracker show that compared with the optimal model in the comparison experiment, the TRGCN model has an average increase of 4.16% in Hits@100 metric and 13.26% in map@100 metric. The validity and rationality of the model are proved.
Most of the existing hypernetwork representation learning methods fail to fully consider the hyperedges, leading to the untapped potential of information contained within the hyperedges. To address this issue, this article proposes a heterogeneous hypernetwork representation learning method with hyperedge fusion abbreviated as HRHF. First, this method incorporates the hyperedges into random walk node sequences by means of incidence graph to enhance tuple relationships, i.e., the hyperedges among the nodes. Second, under the condition of the above random walk node sequences, the cognitive structure model, cognitive set model, and cognitive hyperedge model are jointly optimized to comprehensively consider pairwise relationships and tuple relationships among the nodes to learn high-quality node representation vectors. The experimental results demonstrate that, for the link prediction task, this method outperforms other optimal baseline methods, i.e., hyper-path- based random walks + hyper-gram (HPHG) by 0.99% points on the drug dataset, is comparable to the other optimal method, i.e., Event2vec on the global positioning system (GPS) dataset, is close to the performance of other optimal methods, i.e., hyper-path-based random walks + skip-gram (HPSG) on the MovieLens, and is close to the performance of other optimal methods, i.e., Hyper2vec on the WordNet dataset. For the hypernetwork reconstruction task, this method achieves superior average performance on the drug and GPS datasets compared with other baseline methods.
文本分类任务通常依赖足量的标注数据,针对低资源场景下的分类模型在小样本上的过拟合问题,提出一种基于提示学习的小样本文本分类方法BERT-P-Tuning.首先,利用预训练模型BERT(Bidirectional Encoder Representations from Transformers)在标注样本上学习到最优的提示模板;然后,在每条样本中补充提示模板和空缺,将文本分类任务转化为完形填空任务;最后,通过预测空缺位置概率最高的词并结合它与标签之间的映射关系得到最终的标签.在公开数据集FewCLUE上的短文本分类任务上进行实验,实验结果表明,所提方法相较于基于BERT微调的方法在评价指标上有显著提高.所提方法在二分类任务上的准确率与F1值分别提升了25.2和26.7个百分点,在多分类任务上的准确率与F1值分别提升了6.6和8.0个百分点.相较于手动构建模板的PET(Pattern Exploiting Training)方法,所提方法在两个任务上的准确率分别提升了2.9和2.8个百分点,F1值分别提升了4.4和4.2个百分点,验证了预训练模型应用在小样本任务的有效性.
多模态医学图像可以为临床医生提供靶区(如肿瘤、器官或组织)的丰富信息.然而,由于多模态图像之间相互独立且仅有互补性,如何有效融合多模态图像并进行分割仍是亟待解决的问题.传统的图像融合方法难以有效解决此问题,因此基于深度学习的多模态医学图像分割算法得到了广泛的研究.从原理、技术、问题及展望等方面对基于深度学习的多模态医学图像分割任务进行了综述.首先,介绍了深度学习与多模态医学图像分割的一般理论,包括深度学习与卷积神经网络(CNN)的基本原理与发展历程,以及多模态医学图像分割任务的重要性;其次,介绍了多模态医学图像分割的关键概念,包括数据维度、预处理、数据增强、损失函数以及后处理等;接着,对基于不同融合策略的多模态分割网络进行综述,对不同方式的融合策略进行分析;最后,对医学图像分割过程中常见的几个问题进行探讨,并对今后研究作了总结与展望.
针对SM4门限实现(TI)面积大、随机数消耗多的问题,提出一种SM4门限实现的改进方案.在满足门限实现理论的情况下,对S盒非线性求逆进行了无随机共享,并引入面向域的乘法掩码方案,将S盒随机数消耗减少至12 bit;基于流水线思想,设计了新的8 bit数据位宽的SM4串行体系结构,复用门限S盒,并优化SM4线性函数,使SM4门限实现面积更加紧凑,仅6 513 GE,相较于128 bit数据位宽的SM4门限实现方案,所提方案的面积减小了63.7%以上,并且更好地权衡了速度和面积.经侧信道检验,所提出的改进方案具备抗一阶差分功耗分析(DPA)能力.
Small object detection is one of the important tasks in object detection. Compared with the whole image, the small object has a low proportion, and the feature information of the object is very little, which brings a huge challenge to small object detection. In order to address the issue, we propose a small object detection algorithm based on split mixed attention, named as SMAMYOLO. Firstly, based on the channel attention module and the spatial attention module, we recombine their connection structures and propose a mixed attention module, named as MAM which can enhance the model’s ability to express small object features in spatial dimension. Secondly, according to the effects of different receptive field sizes on the object, we propose a split feature extraction module based on mixed attention, named as SMAM which can adaptively adjust the size of the receptive field according to the scale of the input feature map, and use mixed attention to enhance the ability of different branches to capture small object feature information.Finally, based on SMAM and CSPNet, the core residual module in yolov5 is improved, and a feature extraction module, named as CSMAM, is proposed. The additional computational overhead of CSMAM can be ignored. Based on the TinyPerson dataset, the experiments have been conducted. Firstly, the ablation experiment shows that the average detection accuracy of SMAM-YOLO algorithm is improved by 4.15% compared with the baseline algorithm, and the FPS reaches 74 frame/s; In addition, compared with some existing mainstream small target detection models, the experimental results show that the SMAM-YOLO algorithm improves the detection accuracy by 1.46%～6.84% on average, and can meet the requirements of real-time detection.
基因型填补可以通过填补估算出在基因测序数据中未覆盖的样本区域弥补因技术限制导致的缺失,但现有的基于深度学习的填补方法不能有效捕捉到全序列位点间的连锁关系,造成整体填补准确率低、批量序列填补准确率分散等问题.针对这些问题提出一种融合卷积与自注意力机制的填补方法——FCSA,使用两种融合模块构成编解码器组建网络模型.编码器融合模块使用自注意力层得到全序列位点间的关联度,将该关联度融合到全局位点后再通过卷积层提取局部特征;解码器融合模块使用卷积对编码后的低维向量进行局部特征重建,应用自注意力层对全序列建模并融合.使用多物种的动物基因数据进行模型训练,并在Dog、Pig和Chicken数据集上进行比较验证,结果表明,与SCDA(Sparse Convolutional Denoising Autoencoders)、AGIC(Autoencoder Genome Imputation and Compression)和U-net相比,FCSA在10%、20%和30%缺失率下的平均填补准确率均取得了最高值,且批量序列填补准确率的分散程度较小;消融实验的结果也表明,这两种融合模块的设计能够有效提升基因型填补的准确率.
近年来,网格化分布式新安江模型(GXM)在洪水预报中发挥了重大作用,但在进行洪水过程模拟时,模型数据量与计算量巨大,GXM的计算时间随着模型预热期的增加呈指数增长,严重影响GXM的计算效率.因此,提出一种基于网格流向划分与动态优先级有向无环图(DAG)调度的GXM并行算法.首先,对模型参数、模型构件、模型计算过程进行分析;其次,从空间并行性的角度提出了基于网格流向划分的GXM并行算法以提高模型的计算效率;最后,提出一种基于动态优先级的DAG任务调度算法,通过构建网格计算节点的DAG并动态更新计算节点的优先级以实现GXM计算过程中的任务调度,减少模型计算中数据倾斜现象的产生.在陕西省大理河流域与安徽省屯溪流域对提出的算法进行实验,在预热期为30 d、数据分辨率为1 km的情况下,相较于传统的串行算法,所提算法的最大加速比分别达到了4.03和4.11,有效提升了GXM的计算速度与资源利用率.
网络中存在大量语义相同或者相似的冗余文本,而文本去重能够解决冗余文本浪费存储空间的问题,并能为信息抽取任务减少不必要的消耗.传统的文本去重算法依赖文字重合度信息,而没有较好地利用文本语义信息,同时也无法捕捉长文本中距离较远句子之间的交互信息,去重效果不理想.针对文本语义去重问题,提出一种基于关键词图表示的长文本去重算法.首先,通过抽取文本对中的语义关键词短语,将文本对表示为以关键词短语为节点的图;其次,通过多种方式对节点进行编码,并利用图注意力网络(GAT)学习节点之间的关系,从而得到文本对图的向量表示,并判断文本对是否语义相似;最后,根据文本对的语义相似度进行去重处理.与传统算法相比,所提算法能有效利用文本的语义信息,并能通过图结构将长文本中距离较远的句子用关键词短语的共现关系连接起来,从而增加不同句子之间的语义交互.实验结果表明,所提算法在两个公开数据集CNSE(Chinese News Same Event)和CNSS(Chinese News Same Story)上都取得了比Simhash、BERT(Bidirectional Encoder Representations from Transformers)微调、概念交互图(CIG)等传统算法更好的表现.具体地,所提算法在CNSE数据集上的F1值达到84.65%,在CNSS数据集上的F1值达到90.76%,说明所提算法可以有效提升文本去重任务的效果.
The aim of Network Representation Learning（NRL） is to learn the potential and low-dimensional representation of network nodes, and the low-dimensional representation is applied to network analysis for downstream tasks. The present NRL algorithms using autoencoder insufficiently extracted information about node attributes and generated information bias, which affected the effectiveness of learning. In the proposed model, Network Representation learning model based on Autoencoder with optimized Graph Structure（NR-AGS） was proposed to improve the accuracy by optimizing the graph structure. Firstly, the structure and attribute information were fused to generate the joint transfer matrix which formed high-dimensional representation. Secondly, the low-dimensional embedded representation was gained by autoencoder. Finally, the deep embedded clustering algorithm was introduced to form a self-supervision mechanism in the process of the training of autoencoder and the category distribution division of nodes. In addition, the improved Maximum Mean Difference（MMD） algorithm was introduced to reduce the gap between distribution of low-dimensional embedded representation and distribution of original data. The proposed model was gained using the reconstruction loss of the autoencoder, the deep embedded clustering loss and the improved MMD loss. Three real datasets were trained by the NR-AGS model. The obtained low-dimensional representation was applied to node classification and node clustering. Compared with the deep graph representation model Deep Neural networks for Graph Representations（DNGR）, NR-AGS improves by 8.3, 19.8and 9.6percentage points on Cora, Citeseer and Wiki datasets. The experimental results show that NR-AGS model can greatly improve the learning outcomes of network representation learning.
To address the problem of low detection accuracy, inability to adapt to environmental changes and noise effects for traditional machine vision methods in surface defect detection, an improved deep learning surface defect detection algorithm based on wavelet transform was proposed, namely WT-U-Net++. Firstly, the low frequency and high frequency components of the defect image were obtained by the wavelet transform, and then the detailed features of the high and low components were extracted by the multi-scale module Mix-conv Inception（MCI）. Then, the detailed features extracted by MCI were fused with the original image as the input of the improved U-Net++. In the downsampling stage of U-Net++, channel attention module was introduced to enable the network to capture more contextual semantic information and strengthen the quality of cross-layer features cascade. In the upsampling stage, deconvolution was adopted to recover more defect details. Finally, the best result was selected from the multiple outputs of U-Net++ as the detection result. Two comprehensive indexes of Intersection-over-Union ratio（IoU） and Dice Similarity Cofficient（DSC） were evaluated experimentally. Experimental results on three public defect data sets of rail, magnetic tile and silicon steel oil stain shown that, compared with the sub-optimal algorithm U-Net ++, WT-U-Net ++ has the IoU improved by 8.00%, 4.63%, and 8.74%, and has DSC improved by 4.26%, 2.99% and 4.64%.
在传统多背包问题的基础上,从典型物流服务场景中共性抽象出异构多背包问题(HMKP),并设计和定制了一种帝国竞争算法(ICA)对HMKP进行求解和评估.针对原始ICA易陷入局部最优以及0-1背包问题最优解往往在约束边界周围的特点,设计了双点自变异策略(TPAS)和跳出局部最优算法(JLOA)对ICA进行改进,提出面向0-1背包问题的二进制帝国竞争算法(BICA).BICA在求解35个0-1背包问题算例时展现出了全面、高效的寻优能力,基于最佳匹配值法(BMV)的BICA在第一组测试集的20个算例上能对19个算例100%找到理想最优值,在第二组测试集的15个算例上能对12个算例100%找到理想最优值,在所有对比算法中表现最优.数值结果分析表明,BICA在寻优演化中维持多极发展策略,并依托独特的种群进化方式在解空间中高效搜索理想解.在此基础上,针对HMKP强约束性和高复杂度的特性,基于BICA设计了求解HMKP的多级二进制帝国竞争算法(MLB-ICA).分别在多个典型0-1背包问题算例组合构建的HMKP高维测试集上进行了MLB-ICA的数值实验和性能评估,结果表明虽然MLB-ICA的求解时间比Gurobi长,但求解精度提高了28%.可见,MLB-ICA能以较低的计算代价在可接受的时间范围内高效求解高维复杂的HMKP,为ICA在超大规模组合优化问题中的求解提出了可行的算法设计方案.
To address the issues of blurred contour and lost detail of portrait image after restoration owing to motion blur, a method of generation adversarial network based on multi-stage jump residual group for moving portrait deblurring was proposed. Firstly, the residual block was refined to construct the multi-stage jump residual group module and then the structure of PatchGAN was also improved,which made Generated Adversarial Network（GAN）better combining with the image features of each layer. Secondly, the multi-loss fusion method was adopted to optimize the network, which can enhance the real texture of the reconstructed image. Finally, the end-toend mode was advocated to blind deblurring blurred portrait image and output clear portrait image. Experimental results on CelebA dataset show that Peak Signal to Noise Ratio（PSNR） and Structural Similarity（SSIM） of the proposed method are at least 0.46 dB and 5.6% higher than those of the CNN-based methods such as DeblurGAN and SRN-DeblurNet（Scale-current Network）. This method has fewer model parameters, faster restoration speed, and more texture details in the restored portrait image.
现实生活中的多目标优化问题(MOP)大多为动态多目标优化问题(DMOP),此类问题的目标函数、约束条件和决策变量都可能随时间的变化而发生改变,这需要算法在环境变化后快速适应新的环境,且在保证Pareto解集多样性的同时快速收敛到新的Pareto前沿.针对此问题,提出一种基于新评价指标自适应预测的动态多目标优化算法(NEI-APDMOA).首先,在种群非支配排序过程中提出一种优于拥挤度的新评价指标,并分阶段平衡收敛快速性和种群多样性,使种群的收敛过程更加合理;其次,提出一种可判断环境变化强弱的因子,为预测阶段提供有价值信息,并引导种群更好地适应环境变化;最后,根据环境变化因子匹配3种更加合理的预测策略,使种群快速响应环境变化.将NEI-APDMOA与DNSGA-Ⅱ-A(Dynamic Non-dominated Sorting Genetic Algorithm-Ⅱ-A)、DNSGA-Ⅱ-B(Dynamic Non-dominated Sorting Genetic Algorithm-Ⅱ-B)和PPS(Population Prediction Strategy)算法在9个标准动态测试函数上进行对比.实验结果表明,NEI-APDMOA分别在9、4和8个测试函数上取得了最优的平均反世代距离(IGD)值、平均间距(SP)值和平均世代距离(GD)值,可以更快地响应环境变化.
边缘计算将计算资源部署在离终端用户更近的边缘计算节点,从待选的位置中选择合适的边缘计算节点部署位置能提升边缘计算服务的节点容量以及用户服务质量(QoS).然而,目前对于如何放置边缘计算节点以降低边缘计算成本的研究较少.此外,在边缘服务的时延等QoS因素的约束下,目前尚没有一种边缘计算节点部署算法能最大限度地提高边缘服务的鲁棒性同时最小化边缘节点部署成本.针对上述问题,首先,通过建立计算节点、用户传输时延和鲁棒性的模型将边缘计算节点放置问题转化为带约束条件的最小支配集问题;随后,提出重合支配的概念,基于重合支配衡量网络鲁棒性,设计了基于重合支配的边缘计算节点放置算法——CHAIN(edge server plaCement algoritHm based on overlApping domINation).仿真实验结果表明,与面向覆盖的近似算法和面向基站的随机算法相比,CHAIN的系统时延降低了50.54%与50.13%.
句向量是自然语言处理的核心技术之一,影响着自然语言处理系统的质量和性能.然而,已有的方法无法高效推理句与句之间的全局语义关系,致使句子在欧氏空间中的语义相似性度量仍存在一定问题.为解决该问题,从句子的局部几何结构入手,提出一种基于流形学习的句向量优化方法.该方法利用局部线性嵌入(LLE)对句子及其语义相似句子进行两次加权局部线性组合,这样不仅保持了句子之间的局部几何信息,而且有助于推理全局几何信息,进而使句子在欧氏空间中的语义相似性更贴近人类真实语义.在7个文本语义相似度任务上的实验结果表明,所提方法的斯皮尔曼相关系数(SRCC)平均值相较于基于对比学习的方法SimCSE(Simple Contrastive learning of Sentence Embeddings)提升了1.21个百分点.此外,将所提方法运用于主流预训练模型上的结果表明,相较于原始预训练模型,所提方法优化后模型的SRCC平均值提升了3.32～7.70个百分点.
虚假招聘广告的泛滥不仅会损害求职者的合法权益,还会扰乱正常的就业秩序,造成求职者极差的用户体验.为了有效检测出虚假招聘广告,提出一种基于一致性训练的半监督虚假招聘广告检测模型(SSC).首先,对所有数据应用一致性正则项提升模型的性能;然后,通过联合训练的方式整合有监督损失和无监督损失得到半监督损失;最后,使用半监督损失对模型进行优化.在两个真实数据集EMSCAD(EMployment SCam Aegean Dataset)和IMDB(Internet Movie DataBase)上的实验结果表明,SSC在标签数据仅为20时取得了最好的检测效果,准确率与现有先进的半监督学习模型UDA(Unsupervised Data Augmentation)相比提升了2.2和2.8个百分点,与深度学习模型BERT(Bidirectional Encoder Representations from Transformers)相比提升了3.4和11.7个百分点,同时还具有较好的可拓展性.
