Existing research on multi-QoS scheduling problems,due to its reliance solely on immediate reward feedback mechanisms,faces issues of poor scalability and resource wastage when handling delay-sensitive data and media data with con-tinuous transmission requirements in resource-constrained scenarios.To address this problem,this paper proposed a RB-DQN algorithm.This algorithm adjusted the current state's policy evaluation by backtracking future interactions,effectively identif-ying and resolving packet loss caused by suboptimal scheduling strategies.Additionally,it designed a LTT metric,which com-prehensively considered the service requirements of both delay-sensitive data and media-type data,allowing for weight adjust-ments to emphasize different priorities.Extensive simulation results demonstrate that the proposed algorithm significantly re-duces the delay and jitter of delay-sensitive data while ensuring the smoothness and stability of media-type data,outperforming other scheduling strategies.
The development of the internet and mobile technologies has driven the connection of devices such as smartphones to the network, resulting in a multitude of scattered data points. Restrictions on network bandwidth and the risk of privacy leaks have limited the widespread application of centralized machine learning. In contrast, federated learning technology, with its privacy- preserving feature, optimizes data processing by utilizing distributed network resources. Although federated learning reduces the demand for data transmission through parameter sharing, the exposure of parameters during network transmission still poses a risk to privacy security. In this study, a robust and privacy-preserving federated learning scheme is proposed, which performs user selection based on ciphertext. Firstly, the concept of user groups is constructed on the edge server, and a ciphertext-based user selection algorithm is designed to select users with large and evenly distributed datasets to participate for the training of the global model within the framework of Federated Learning. In addition, the algorithm effectively filters out anomalous gradients that significantly differ from the global model parameters while ensuring that the privacy of users' local model parameters is not compromised, thereby enhancing the overall robustness of the global model. Secondly, a lightweight encryption algorithm is designed based on cloud-edge collaboration, which not only effectively protects the privacy of users' local model parameters but also ensures the security of these parameters during the aggregation process, with high computational and communication efficiency. Based on the difficulty of the Decisional Diffie-Hellman (DDH) problem, the proposed scheme has been proven to be secure and effectively resistant to various attacks, ensuring the robustness of the global model and protecting users' data privacy. Compared with existing schemes, performance analysis and experimental results show that the proposed scheme performs better in terms of efficiency and accuracy.
The inspection of aircraft skin is critical to ensuring the safety of aircraft flights.Employing autonomous mobile ro-bots for inspection can significantly enhance inspection efficiency and reduce safety risks.However,achieving full coverage of aircraft skin with a single type of robot is challenging due to the complexity of aircraft structures.Therefore,this paper pro-posed an aerial-ground cooperative coverage path planning(AG-CCPP)method for heterogeneous robots.Firstly,this paper introduced a heterogeneous robot system consisting of unmanned aerial vehicles(UAV)and unmanned ground vehicles(UGV),analyzed the necessary constraints in the planning process,including workspace constraints,endurance constraints,and so on,and established an optimization model using integer linear programming.Subsequently,this paper proposed a dual-chromosome genetic algorithm with a multi-elite population based on a greedy allocation strategy for jointly solving task alloca-tion and path planning.By incorporating an allocation chromosome,it achieved joint solution for task allocation and path plan-ning,enabling global optimization.It conducted greedy allocation based on endurance constraints,fully leveraging the advan-tages of heterogeneous robots.It adopted a multi-level elite population design to reduce the number of inefficient crossover pop-ulations and enhance the algorithm's operational efficiency.Finally,simulation experiments and comparative analysis conduc-ted on a Boeing 737-300 demonstrate that the proposed method outperforms existing algorithms in terms of both the completion time for cooperative coverage by robots and the program execution time.
To address the problems of malicious node aggregation,high shard failure rate,and load imbalance in the current models of blockchain shard,this paper proposed a blockchain dynamic sharding model(CBDSM)based on node credibility.Firstly,this paper constructed a blockchain sharding architecture to solve the scalability bottlenecks faced by the blockchain.Secondly,it designed the reliability evaluation mechanism of nodes to quantify the ability and reliability of nodes,which rea-lized the efficient screening and classification of nodes.On this basis,it designed the dynamic network fragmentation algorithm and the network fragmentation flow.Through the dynamic allocation of nodes at different levels,it realized load balancing on the chain.Theoretical analysis and experimental results show that,compared with similar models,CBDSM boasts a 54％in-crease in throughput,a 46％reduction in consensus failure rate,and a 15.6％decrease in shard failure rate.This model can guarantee the dynamic equilibrium of nodes,and resist the single-point aggregation and collusion attacks of nodes.It ensures the dynamic security of the blockchain shard system,with the high performance and high scalability of the system.
Log parsing, the process of transforming raw logs into structured data, is a key step in the complex computer system's intelligent operation and maintenance and therefore has received extensive attention. Among all log parsing methods, heuristic log parsing methods are lightweight and can work in a streaming mode to well meet the real-time parsing requirements. However, the existing log representations used in the heuristic log parsing methods are not powerful in distinguishing log messages, which leads to low parsing accuracy and weak generality. Inspired by trigger word extraction of the event detection task in natural language processing (NLP), this paper proposes an online log parser, named PosParser, which employs the part-of-speech (PoS) tagging to extract a function token sequence (FTS) as the log message representation, and then identify event templates of log messages through the FTS. Experimental results on sixteen logs from real systems demonstrate that the FTS is powerful in distinguishing log messages from different event templates, and PosParser not only performs better in terms of parsing accuracy than state-of-the-art methods but is also comparable to them in efficiency.
Medical images have low contrast and blurred boundaries between different tissues or between tissues and lesions. Because labeling medical images is laborious and requires expert knowledge, the labeled data are expensive or simply unavailable. UNet has achieved great success in the field of medical image segmentation. However, the pooling layer in downsampling tends to discard important information such as location information. It is difficult to learn global and long-range semantic interactive information well due to the locality of convolution operation. The usual solution is increasing the number of datasets or enhancing the training data though augmentation methods. However, to obtain a large number of medical datasets is tough, and the augmentation methods may increase the training burden. In this work, we propose a 2D medical image segmentation network with a convolutional capsule encoder and a multiscale local co-occurrence module. To extract more local detail and contextual information, the capsule encoder is introduced to learn the information about the target location and the relationship between the part and the whole. Multi-scale features can be fused by a new attention mechanism, which can then selectively emphasize salient features useful for a specific task by capturing global information and suppress background noise. The proposed attention mechanism is used to preserve the information that is discarded by pooling layers of the network. In addition, a multi-scale local co-occurrence algorithm is proposed, where the context and dependencies between different regions in an image can be better learned. Experimental results on the dataset of Liver, ISIC and BraTS2019 show that our network is superior to the UNet and other previous medical image segmentation networks under the same experimental conditions.
Given the current limitations in intelligence and processing capabilities, machine learning systems are yet unable to fully tackle diverse scenarios, thereby restricting their potential to completely substitute for human roles in practical applications. Recognizing the robustness and adaptability demonstrated by human drivers in complex environments, autonomous driving training has incorporated driving intervention mechanisms. By integrating these interventions into Proximal Policy Optimization (PPO) algorithms, it becomes possible for drivers to intervene and rectify vehicles' irrational behaviors when necessary, during the training process, thereby significantly accelerating the enhancement of model performance. A human-centric experiential replay mechanism has been developed to increase the efficiency of utilizing driving intervention data. To evaluate the impact of driving intervention on the performance of intelligent agents, experiments were conducted across four distinct intervention frequencies within scenarios involving lane changes and navigation through congested roads. The results demonstrate that the bespoke intervention mechanism markedly improves the model's performance in the initial stages of training, enabling it to overcome local optima through timely driving interventions. Although an increase in intervention frequency typically results in improved model performance, an excessively high intervention rate can detrimentally affect the model's efficiency. To assess the practical applicability of the algorithm, a comprehensive testing scenario that includes lane changes, traffic signals, and congested road sections was devised. The performance of the trained model was evaluated under various traffic conditions. The outcomes reveal that the model can adapt to different traffic flows, successfully and safely navigate the testing segment, and maintain speeds close to the target. These findings highlight the model's robustness and its potential for real-world application, emphasizing the critical role of human intervention in enhancing the safety and reliability of autonomous driving systems.
System logs record system operation status and important event information. They are the important basis for debugging system failures and cause analysis. Due to the low accuracy of log parsing and insufficient labeled samples, anomaly detection precision is low. Therefore, we propose a new log-based semi-supervised anomaly detection method named BTCNLog. Firstly, the improved log parsing method with the dictionary keeps part of the parameter information in the log event. So, it can improve the utilization rate of log information and the accuracy of log parsing. Then, BERT is used to encode the semantic information to obtain the semantic vector of the log for the template. What's more, the clustering method is applied to estimate the tag to deal with insufficient data tagging problems. Therefore, it can improve the ability to detect unstable data for the model. Finally, a bidirectional temporal convolution network (Bi-TCN) with residual blocks is introduced to capture contextual information from two directions to improve the accuracy and efficiency of anomaly detection. To evaluate the performance of the proposed method, BTCNLog is compared with six baselines on two datasets. The final experimental results show that, compared with the latest three benchmark models, LogBERT, PLELog, and LogEncoder, the proposed method showed an average improvement of 7%, 14.1%, and 8.04% in F1 values.
To protect the copyright of the 3D scene represented by the neural radiation field, the embedding and extraction of the neural radiation field watermark are considered as a pair of inverse problems of image transformations. A scheme for protecting the copyright of the neural radiation field is proposed using invertible neural network watermarking, which utilizes watermarking techniques for 2D images to achieve the protection of the 3D scene. The scheme embeds the watermark in the training image of the neural radiation field through the forward process in the invertible network and extracts the watermark from the image rendered by the neural radiation field using the inverse process to realize the copyright protection of both the neural radiation field and the 3D scene. Since the rendering process of the neural radiation field can cause the loss of watermark information, the scheme incorporates an image quality enhancement module, which utilizes a neural network to recover the rendered image and then extracts the watermark. The scheme embeds a watermark in each training image to train the neural radiation field and enables the extraction of watermark information from multiple viewpoints. Simulation experimental results demonstrate the effectiveness of the method.
The single domain generalization(SDG) based on meta-learning has emerged as an effective technique for solving the domain-shift problem. However, the inadequate match of data distribution between source and augmented domains and difficult separation of domain-invariant features from domain-related features make SDG model hard to achieve great generalization. Therefore, a novel meta-learning method based on domain enhancement and feature alignment (MetaDefa) is proposed to improve the model generalization performance. First, the background substitution and visual corruptions techniques are used to generate diverse and effective augmented domains. Then, the multi-channel feature alignment module based on class activation maps and class agnostic activation maps is designed to effectively extract adequate transferability knowledge. In this module, domain-invariant features can be fully explored by focusing on similar target regions between source and augmented domains feature space and suppressing the feature representation of non-similar target regions. Extensive experiments on two publicly available datasets show that MetaDefa has significant generalization performance advantages in unknown multiple target domains.
A watermarking algorithm is proposed in this paper to address the copyright protection issue of implicit 3D models. The algorithm involves embedding watermarks into the images in the training set through an embedding network, and subsequently utilizing the NeRF model for 3D modeling. A copyright verifier is employed to generate a backdoor image by providing a secret perspective as input to the neural radiation field. Subsequently, a watermark extractor is devised using the hyperparameterization method of the neural network to extract the embedded watermark image from that perspective. In a black box scenario, if there is a suspicion that the 3D model has been used without authorization, the verifier can extract watermarks from a secret perspective to verify network copyright. Experimental results demonstrate that the proposed algorithm effectively safeguards the copyright of 3D models. Furthermore, the extracted watermarks exhibit favorable visual effects and demonstrate robust resistance against various types of noise attacks.
With the development of electronic technology, intelligent cars can gradually realize more complex artificial intelligence algorithms. The video caption algorithm is one of them. However, current video caption algorithms only consider single-visual information when applied to urban traffic scenes, which leads to the failure to generate accurate captions of complex sets. The multimodal fusion algorithm based on Transformer is one of the solutions to this problem. However, the existing algorithms have the difficulties of a low fusion performance and high computational complexity. We propose a new video caption Transformer-based model, the MFVC (Multimodal Fusion for Video Caption), to solve these issues. We introduce audio modal data and the attention bottleneck module to increase the available information to describe the generative model and improve the model effect with less operation costs through the attention bottleneck module. Finally, the experiment is conducted on the available datasets, MSR-VTT and MSVD. Meanwhile, to verify the effect of the model on the urban traffic scene, the experiment is carried out on the self-built traffic caption dataset BUUISE, and the evaluation index confirms the model. This model can achieve good results on both available datasets and urban traffic datasets and has excellent application prospects in the intelligent driving industry.
由于忽略了对多尺度特征的提取,以及不同层次特征之间的差异,显著性目标检测仍然存在预测结构不完整、细节丢失的问题.为了解决这些问题,提出了一个新的显著性检测模型M3Net.该网络主要由多尺度特征自适应融合模块和循环反馈聚合模块组成.多尺度特征自适应融合模块旨在自适应地捕捉和聚合不同层次的多尺度特征.循环反馈聚合模块组成可以在迭代过程中对不同层次特征聚合的同时,有效地防止特征的稀释.在五个基准数据集上的实验结果表明,该网络在Fβ、Em、MAE三种评价指标上优于十种现有网络.在DUT-OMRON数据集上,Fβ 指标比排名第二的显著性检测模型提高了 0.4％,Em指标提高了 0.3％;在ECSSD数据集上,Fβ指标比排名第二的显著性检测模型提高了 0.2％,Em指标提高了 0.3％,同时网络还具有优秀的速度表现.
如何高效快速地应对网络中的故障是设计路由协议的基本要求和主要任务.由于动态路由协议在应对网络中的故障时,在协议动态收敛的过程中将会有大量的报文被丢弃.因此,目前路由器厂商普遍采用路由保护方法来克服网络故障,在众多的路由保护方法中,DC(downstream criterion)规则是一种被普遍认可的方法.然而,已有的实现DC规则算法的时间复杂度普遍较高,并且复杂度随着网络节点平均度的增加而迅速增加.为了应对上述问题,提出一种线性时间复杂度的高效路由保护方案ERPLR(efficient routing protection method with 1inear time complexity),该方法首先提出了备份下一跳计算规则,然后在已有最短路径树的基础上,根据备份下一跳计算规则为所有的源目的节点对计算备份下一跳.在计算备份下一跳的过程中,每个节点和其邻居最多被访问一次,因此ERPLR的时间复杂度为O(V+E).实验结果表明,与已有的实现DC规则相比较,ERPLR在故障保护率和路径拉伸度两个度量指标结果相似的情况下,在真实网络拓扑和模拟拓扑中,ERPLR分别降低了大约74.93％和78.91％的计算开销.该方法可以极大地降低DC规则的计算开销.
高效的全景视频压缩对沉浸式、低延时的视觉体验十分重要.服务器端的全景视频通过投影到2D平面,经过编码、传输到客户端后,反投影到球面供用户观看,而全景视频投影过程中会在两极产生拉伸形变,由于过采样而产生像素冗余,破坏原有的信息熵分布.针对这一问题,提出一种改进的熵平衡算法,根据全景视频投影前后对应编码单元的熵不变原则,采用码率控制的R-λ模型根据投影面积变化比率来重新分配目标比特,以达到熵平衡,进而调整编码参数优化编码.实验结果表明,该算法相较于基准VTM14.0能达到1.87％的码率节省,编码时间下降10.07％,说明该算法在全景视频编码中能实现更高的压缩率和更短的编码时间.
Federated learning can protect the data privacy of edge devices in collaborative training of edge devices. In the general FL scenarios, the participants of FL are usually composed of heterogeneous edge devices, where resource-constrained devices will consume more time, resulting in the decline of the training speed. The existing schemes either ignore stragglers, or distribute the computing tasks according to the distributed idea, but the distribution process involves the transmission of raw data, which cannot guarantee data privacy. To alleviate the straggler problem in small or medium-sized multiple heterogeneous devices scenario, this paper proposed a coding-based FL scheme, and it designed an efficient scheduling algorithm combined with the mathematical characteristics of linear coding to ensure data privacy and accelerate the speed of heterogeneous FL system. Meanwhile, the experimental results completed in the actual experimental platform show that when the performance diffe-rence between heterogeneous devices is large, the coding-based FL scheme can shorten the training time of the straggler by 92.85%.
Due to the lack of tailored security and privacy technology in many fields such as Internet of Things, artificial intelligence and big data, the system based on distributed ledger technology（DLT） can well meet their the security requirements as a decentralized data management method. Consequently, systems based on distributed ledger technology emerge in an endless stream. In order to help subsequent researchers understand the field of distributed ledger technology more comprehensively and quickly, this paper reviewed the research results based on distributed ledger technology in recent years. It analyzed the blockchain in depth from the perspectives of data structure, technical principle, consensus mechanism and performance, and compared and analyzed some recent advances in distributed ledger technology in detail based on the current research status, such as Hashgraph, Holochain and Truechain. Finally, this paper summarized the future development prospect of this field.
针对传统遗传算法(genetic algorithm,GA)求解旅行商问题(traveling salesman problem,TSP)存在寻优效率低、实验结果缺乏一致性等问题,提出了一种基于基因库的遗传算法(genetic algorithm based on genes pool,GPGA).GPGA从种群中搜索减小哈密顿圈长度的边,并当做优良基因构成基因库.父代哈密顿圈在基因库引导下产生更优的子代哈密顿圈,基因库也随着种群的不断进化而同步更新,引导种群个体逐步向最优解靠近.算例结果表明在同样条件下,GPGA比传统遗传算法和几种改进遗传算法的性能更优.
为更有效地解决以最大完工时间最小化为目标的置换流水车间调度问题,提出了一种自适应混合粒子群算法(SHPSO).该算法结合Q学习设计了参数自适应更新策略,以平衡算法的探索和开发;同时引入粒子停滞判断方法,使用平局决胜机制和Taillard加速算法改进基于迭代贪婪的局部搜索策略,对全局极值进行局部搜索,帮助粒子跳出局部最优.实验结果表明,对比其他四种改进PSO算法,SHPSO算法取得的平均相对百分偏差(RPDavg)至少下降了 83.2％,在求解质量上具有明显优势.
粒子群优化算法因其支配参数少、收敛速度快、易于实现等特点被广泛应用,但是粒子群优化算法存在精度低、容易陷入局部优化的问题.为此提出一种基于双种群交叉学习的粒子群优化算法.在该算法中,整个种群被分为普通子种群和精英子种群.普通子种群采用综合变异机制,该机制通过设置概率参数使普通子种群随机选择朝着优秀粒子的方向或者保持自身方向进行变异,以侧重寻找可能解区域.精英子种群则采用交叉学习机制,将粒子的历史最优和全局最优个体进行交叉生成范例,从而引导粒子对可能解区域进行局部搜索,还提出了一种非线性惯性权重来平衡粒子的全局勘探和局部开发能力.为了验证算法的有效性,在十六个基准问题上进行测试并与其他七种粒子群优化算法变体比较,实验结果表明该算法在求解精度和收敛速度总体排名第一,验证了该算法求解性能优于其他粒子群优化算法变体.
