With the rapid growth of artificial intelligence and big data, concerns about data privacy and security are increasing. Federated learning (FL), a distributed machine learning approach, has gained significant attention. However, due to the heterogeneity of client data and varying collection methods, client-side data often suffers from severe class imbalance, which can degrade model performance and compromise the effectiveness and reliability of FL. In this paper, we propose a class-adaptive balanced FL algorithm with dual contrastive loss. Specifically, in the data preprocessing stage, oversampling operations are used to solve the problem of unbalanced category distribution in training samples, ensuring the learning effect of the model on samples of each category, thereby reducing the degree of model drift. In addition, two contrastive loss functions, intra-class compactness loss and supervised contrastive loss, are designed to optimize the sample features learned by the model in the feature space. By incorporating these two losses, it aims to address the potential overfitting issue resulting from oversampling, thereby improving the classification performance and generalization ability of the model. Extensive experiments confirm that the method proposed in this paper outperforms other state-of-the-art methods in terms of accuracy and scalability.

In massive open online course (MOOC) discussion board, students’ learning experience, reflecting implicit cognitive and affective states, is related to their learning outcomes and course’s completion rates. The majority of researches about learning experience identification in MOOCs depend on post-hoc questionnaires, which may encounter issues such as personal biases, hazy memories, or time constraints, and distribution difficulty in MOOCs. Moreover, learning experience is influenced by students’ interactions during learning but their relationship has not been thoroughly explored. This study aimed to address these issues. Firstly, it proposed an artificial intelligence-based text analysis approach for automatically identifying patterns of learning experiences from the large-scale students’ posts in MOOC discussion board. It had performance advantage in terms of accuracy when compared with the other competing approaches. Secondly, this study defined students’ interactive roles from both social relations and interaction behaviors in MOOC discussion board, and analyzed learning experiences corresponding to the different interactive roles. For students with high participation and low influence in interactions, flow and boredom were prone to happen, while for students with low participation and high influence in interactions, anxiety and apathy were easy to generate. Finally, this study revealed the effect of learning experience on learning achievement with respect to interactive role. For students with high participation characteristics, their learning achievements were less affected by learning experience, while for students less active in interaction, flow was related with good learning achievements. In summary, this study had significant methodological implications for automated learning experience identification. Moreover, this study revealed importance of interactive role in describing the interplay between learning experience and learning achievement, and provided suggestions for the improvement of MOOCs.

Federated learning (FL), as an emerging paradigm in edge intelligence, enables numerous edge devices to collaborate with a central server in training a shared model without compromising data privacy. Existing FL algorithms mostly focus on addressing the problem of non-independently and identically distributed (Non-IID) data in supervised scenarios. However, in practical settings, a considerable amount of unlabeled data exists, posing an urgent challenge of how to fully utilize such data. In this paper, we propose an exploitation maximization method of unlabeled data for federated semi-supervised learning (FSSL). Specifically, we adjust the optimization objective of high-confidence samples by considering both the predicted results and the class distribution of pseudo-labels. This adjustment enables the resolution of severe class imbalance issues. Additionally, we propose a diversity sampling strategy for low-confidence samples. The strategy aims to increase the diversity of training samples by sampling samples that are less similar to high-confidence samples, thereby improving the generalization ability of the model. Extensive experiments demonstrate that our method not only outperforms other state-of-the-art FSSL methods in non-IID scenarios but also exhibits superior scalability and robustness.

Federated learning (FL) as an emerging edge intelligence paradigm allows clients to jointly train a model without exchanging raw data. Due to its excellent performance in privacy protection, FL has practical application in many areas. However, data heterogeneity across clients is a prevalent phenomenon and poses a significant challenge to FL. Although many FL algorithms have been proposed to address the issue of performance deterioration under non-independent and identically distributed (Non-IID), the improvement in performance is not significant. In this paper, we propose knowledge discrepancy-aware federated learning (KDAFL). It evaluates the local model with regard to each class and overall learning effect based on the discrepancy between local and global knowledge. In this way, each client assigns a new weight to each category of cross-entropy and decides if knowledge distillation is to be conducted. The awareness of discrepancy allows the client to adjust the local training according to the characteristics of the knowledge it learned, thus better solving the Non-IID issue. Extensive experiments have demonstrated the effectiveness of KDAFL, particularly in terms of the improvement of global model accuracy compared to other state-of-the-art algorithms.

为了提高弹性光网络的性能,提出一种时频域及阻塞门限联合触发的周期碎片整理(PDT-TFDBT)策略,该策略在触发条件方面采用改进碎片化指数的阻塞延迟触发机制,减少未阻塞时进行无效整理的负载量;在整理顺序方面综合考虑业务的时域和频域特征,合理分配业务的重构顺序,同时引入启发式深度学习辅助的路由及频谱分配(DKA-RSA)算法.仿真结果表明,PDT-TFDBT策略不仅可以减少重构业务数量和频谱碎片,而且进一步降低了业务阻塞率,提高了频谱利用率.

为了有效解决纤芯中的串扰问题并降低网络阻塞率,提出了一种自适应阈值和频谱优先(AT-SF)算法,采用纤芯分组的方式使每组中的纤芯不相邻,将典型的 7 芯光纤分成 3 组,第三组纤芯的优先级在业务到达过程中是可变的;同时,AT-SF算法引入了频隙(FS)阈值参数,将大于FS阈值的业务分配在第一、第二或者第三组纤芯上,小于等于FS阈值的业务只分配在第三组纤芯上.分别在NSFNET、USNET网络中进行了仿真实验,对比了首次匹配(FF)、三维资源分配(3D-RA)、路径优先(aW-PF)、共轭梯度频谱优先(CG-SF)算法性能.仿真结果表明,与其它算法相比,AT-SF算法在网络处于高负载状态时能获得更好的阻塞率和串扰性能.

Engineering education is one of the important parts in current higher education reform, and it is especially pivotal for the information technology field of the country’s major development and strategic industries. Focusing on the meticulous division and rapid technological development in information technology field, along with the cultivation of students’ preliminary ability to solve complex engineering problems, this paper introduces the continuous optimization and improvement of the outcome-based education mechanism in Nanjing University of Posts and Telecommunications (NJUPT). Some innovative initiatives are proposed for undergraduate students’ engineering ability cultivation which empowers students to analyze and solve problems under complex engineering backgrounds.

在基于多芯光纤(MCF)的空分复用弹性光网络(SDM-EON)中,针对纤芯间串扰导致的传输信号物理损伤问题,从频谱分配和纤芯选择角度出发,提出了一种串扰感知的SDM-EON频谱分配算法.该算法在为业务请求分配频谱资源时通过避免填充相邻纤芯来尽量减少单个请求内部的串扰以及请求与请求之间的串扰.数值仿真结果表明:在NSFNet和USNet 2种网络拓扑中,该算法能够在维持较低带宽阻塞率水平的同时有效改善串扰问题.

传统空分复用弹性光网络中仅考虑纤芯频谱资源的无保护分配,缺少对业务生存性的保障.提出了一种基于共享风险组的生存性路由纤芯频谱分配(RCSA)改进算法,该算法基于共享风险纤芯组的设想,在单根多芯光纤中同时分配业务的工作频隙和保护频隙,并基于纤芯间串扰机理引入了业务分配优先度参数,根据优先度选择引起串扰最小的频隙分配方案.仿真结果表明,所提算法可以在保障业务生存性的同时有效减少带宽阻塞率和芯间串扰.

The space division multiplexing based elastic optical network (SDM-EON) can flexibly allocate bandwidth according to the volume of service requirements. And multiple core fiber support great increasements of network capacity simultaneously. Considering the physical damage of the transmitted signal in multi-core fibers (MCF) due to the crosstalk among different cores, it is one of the most important problems for application of SDM-EON. In order to effectively solve the problem of crosstalk, this paper proposes an improved strategy that the services occupying fewer frequency slots can be preferentially allocated to the intermediate core when the traffic load is high, and a modified spectrum allocation algorithm based on frequency slot threshold reduction of crosstalk perception. Theoretical analysis and simulation results proves the improved algorithm can obtain better blocking probability and crosstalk performance.

Routing and spectrum assignments (RSA) is a critical issue for Elastic Optical Networks (EONs). Traditional K-Shortest Path (KSP) algorithm cannot provide optimal routing strategy for dynamic arriving services in complex network environment. An improved heuristic deep-learning assisted algorithm is introduced to achieve deep-learning based RSA strategy (D-RSA) to find the global optimal solution, which can improve the network performance in blocking rate, spectrum utilization and spectrum fragmentation. Simulation results show that the proposed algorithm can improve the blocking rate and spectrum utilization by about 3% and 4% respectively, while reducing the spectrum fragmentation to some extent compared to D-RSA.

Convolutional neural network based network traffic classification scheme suffers many disadvantages such as the complex structure designing, gradient declines or even explodes, the deterioration of prediction accuracy, and etc. A residual network based improved traffic classification algorithm is proposed. The convolution layer and pooling layer in the traditional convolutional neural network are replaced by the residual network layer, which can alleviate the problem that the traditional convolution network is too deep to train effectively. The data feature information learned by the proposed algorithm in the training stage is more comprehensive, and the trained model can also be more accurate. Simulation results show that the improved algorithm has higher accuracy than the traditional neural network, which can be improved from 92.05%to 96.18%.

弹性光网络(EON)中的传统路由频谱分配(RSA)算法多考虑路由跳数或频谱资源占用情况,缺乏时域与相邻链路的信息有效利用.提出一种结合预测的多维感知RSA算法,对持续时间已知业务的历史时间信息通过后向传播神经网络预测未来业务的时间信息,在路由时综合考虑时间、频谱和相邻链路资源占用程度.仿真结果表明:与传统RSA算法相比,多维感知RSA算法能有效降低带宽阻塞率.

The current advanced wireless communication technology has brought new development opportunities for many industries, and practical teaching is an important way to cultivate high-quality communication professionals. Virtual simulation provides students with a good way to master the working principles of the entire wireless communication network. Based on virtual simulation platform, the course team has carried out reform and exploration of experimental teaching, and designed the experimental goal of the entire wireless communication network. Based on the constructivist learning theory, practicing the "four factors" in the teaching. The teaching survey data show that the methods used not only stimulate students' enthusiasm for the experimental courses of wireless communication networks, but also help cultivate students' versatility.

Modern communication technology is bringing new development opportunities for many fields with strong network capabilities, and there is an urgent demand for communication professionals. Practical teaching is an important method for cultivating innovative talents for engineering majors. Based on the virtual simulation platform, this paper designs some wireless communication network experiments, including network architecture experiments: wireless access network, core network and bearer network deployment; wireless communication network key technologies experiments: network slicing, wireless resource management and beam management. Each experiment is designed with experimental content, experimental requirements, capability requirements and personalized requirements. The experimental results demonstrate that the designed experiments are helpful to the cultivation of students' innovative ability while testing the theory.

With the development of educational technology, more and more people are learning and gaining ability through online courses. The excessive number of courses has brought about the problem of information overload, making it necessary to recommend suitable courses for students in online interactive way. Some traditional course recommendation schemes ignore or only roughly exploit the semantics of courses, which may result in sub-optimal recommendation. Moreover, most course recommendation schemes didn't take into account the relationships between courses (e.g., two courses are taught by a teacher, etc.). To solve the above issues, this article proposes a semantic and relationship-aware online course recommendation scheme, SRACR, to recommend favorite courses for students. Specifically, Latent Dirichlet Allocation (LDA) is used to extract the fine-grained semantics of each course represented as the topic vector of the course, and then knowledge graph embedding is adopted to map the course relationships to the course knowledge vector. Then, the course feature vector is obtained through combining the course topic vector and knowledge vector. Finally, treating the feature vector of courses as context, we use a contextual multi-armed bandit-based algorithm to estimate students' preference and recommend courses to students through balancing exploration and exploitation. The experiment results on real educational dataset demonstrate the effectiveness of our proposed method.

弹性光网络(EON)中的业务频谱分配需要同时满足子载波频谱一致性和连续性约束,不同带宽需求的业务分配或释放后会产生大小不同的频谱碎片,影响后续业务的分配和频谱资源利用率.提出一种改进的基于资源节约策略(RSS)的EON碎片整理算法.进行碎片整理时,能够使得整理出来的连续可用频隙数与到达业务所需频隙数尽可能相等,避免产生新的频谱碎片.理论分析和仿真结果表明:在100～600 Er1内,改进的碎片整理算法与传统碎片整理算法相比,带宽阻塞率最优可降低8.54％.

人工智能(AI)技术的飞速发展与广泛应用,使得身处时代变革大潮中的学生们在学习观念、习惯、方法、行为方式等方面均发生了重大变革,同时也对高等教育理念、教学方法、教学模式、评价体系、管理体制等产生了巨大冲击.在紧抓人工智能时代信息素养内涵基础上,针对现有人才培养体系中存在的问题,着重从专业课程体系重构、师资队伍建设、教学模式改革、工程素养培养等四个方面出发,对"人工智能+X"时代电子信息类复合型人才培养模式进行了探索与研究.

As online e-learning systems become more prevalent, there is a growing need for them to accommodate individual differences among students. According to the concept of zone of proximal development (ZPD), it is imperative to provide online students with educational contents that are neither too easy nor too difficult, but are slightly beyond their current abilities. However, following ZPD rule is challenging in online e-learning system, due to the following reasons: the system does not know a priori the ability of the online students, especially for the newly arrived student; the exact relationship between student feedback on teaching and their abilities (i.e., reward/gain function) is extremely complicated, and even unknown to each student. Aiming at solving the issue above, this paper proposes a personalized educational scheme to students, POEM, in order to maximize their accumulative learning gains over multiple rounds. Specifically, instead of assuming any specific formal reward function, we first estimate any unknown reward function from noisy samples using Gaussian process (GP) model. Then, the multi-arm bandit based algorithm is used to select the teaching content with the adaptive difficulty level to balance the effect of exploration and exploitation. The simulation results demonstrate the effectiveness of our proposed method.

Traditional Visible Light Communication (VLC) based indoor positioning algorithm needs long time for positioning processing. An improved k-means method is presents in this paper which allows the positioning terminal to receive the signal strength information generated by different LEDs, and the improved k-means method is implemented to establish the fingerprint database. The triangle positioning method will be firstly used to roughly determine the position, and then adopt the fingerprint matching method for accurate locating. Simulation results show that compared with the traditional fingerprint-based method, the improved k-means method has improved the computing efficiency of positioning process.