Although supervised contrastive learning has achieved success in long-tailed medical image classification, existing methods suffer from insufficient learning of tail classes and bias towards head classes, resulting in biased diagnostic models. This paper proposes a Class-aware Augmentation Contrastive Learning (CACL) method for long-tailed medical image classification, which enriches the information of each class by feature augmentation. Firstly, the distribution of the augmented anchors of a sample anchor is modeled as a Gaussian distribution centered at that sample anchor. To simplify the process of sampling augmented anchors via Gaussian distribution, we derive an upper-bound loss function of the contrastive loss, called balanced implicit augmentation contrastive loss (BIACL). Secondly, we propose the Balanced Hybrid Contrastive Loss (BHCL) to mine the additional information between sample anchors and learnable category prototype, and balance the contribution of all classes and ensure the fairness of learning. Finally, to enhance the classification performance on both head and tail classes thereby efficiently aiding medical diagnosis, we propose an Adaptive Re-balanced Cross-Entropy loss (ARCE). Extensive experiments demonstrate that the proposed CACL outperforms the existing state-of-the-art methods on three benchmark datasets.

Objective. The study aims to tackle optical aberrations induced by sample non-uniformity or imaging system deviations in Fourier Ptychographic Microscopy (FPM). This technique enables high-resolution microscopic imaging with a wide field of view. The proposed approach involves employing a multi-feature fusion residual network method for aberration correction. This method leverages residual networks and multi-scale convolutional kernels to extract deep residual features and enhance the capacity to capture pertinent image characteristics. Approach. This method optimizes network performance by utilizing grouped convolutions and attention mechanisms without significantly increasing the parameter count. Through the combination of Zernike modes, the accurate estimation and correction of aberrations are achieved, resulting in a substantial enhancement in imaging quality. Main results. The research shows the effective correction of aberrations using this method, facilitating high-resolution image reconstruction while preserving intricate texture details. Validation results, derived from simulated and real images, highlight the efficacy of the method. Significance. By introducing the multi-feature fusion residual network method, this study addresses a significant challenge posed by aberrations in FPM, enhancing imaging quality and presenting a new effective approach for high-resolution microscopic imaging. These results have important implications for fields such as biomedicine, which require stringent image quality standards.

Anycast routing enables dynamic destination selection for data flows, improving failover and load balancing in applications such as domain name systems (DNS) and content delivery networks (CDNs). However, existing reinforcement learning (RL) approaches often rely on convolutional neural networks (CNNs) that are poorly matched to non-Euclidean network topologies, neglect the strong coupling between destination selection and path planning, and offer limited support for efficient online interaction. These limitations lead to high-dimensional action spaces, slow convergence, and substantial computation-induced decision latency. To address these challenges, we propose PHGRL-AR, a parallel hierarchical graph reinforcement learning-based anycast routing policy. First, we build an online interactive routing control framework under a software-defined networking (SDN) architecture to enable global dynamic state measurement and reduce computation-induced decision latency. Second, we employ hierarchical RL to decouple destination selection from path planning, thereby reducing action-space dimensionality. During path generation, we introduce an action-masking mechanism based on adjacency relations and shortest-path ordering to prevent invalid actions and routing loops, improving interaction efficiency. Third, to overcome CNN limitations, we incorporate graph RL and design an edge-node switching graph convolution structure. This design aligning with the joint node-link attribute characteristics of anycast routing to enhance generalization. Finally, we parallelize environment interactions during training to reduce computational overhead. Experimental results show that PHGRL-AR significantly reduces end-to-end delay compared with the shortest-shortest-path (SSP) algorithm and a hierarchical graph reinforcement learning baseline, achieving a maximum reduction of 50.3%.

In both resource-scarce and resource-abundant scenarios (i.e., in the scenarios where service demands exceed/do not exceed available resources), it is a critical challenge to design efficient task allocation model and algorithm in network slicing in 5G/Beyond 5G (B5G)/6G networks, especially in resource-scarce scenario. This challenge faces the following three difficult tasks: 1) fully utilize network resources, 2) satisfy all user requests, and 3) make the best trade-offs among service latency, service failure rate, and Mobile Network Operator (MNO) cost. To address this challenge, we propose a three objective optimization model, the Versatile Storage Routing (VSR) model. It facilitates collaborative multi-device sharing for a single Virtual Network Function (VNF) and parallel multi-path data transmission, thus enabling full utilization of the resources. Also, it can provide efficient resource allocation schemes in both resource-abundant and resource-scarce scenarios to satisfy all user requests. Furthermore, we design a novel evolutionary algorithm for VSR model called VSR-EA based on Non-dominated Sorting Genetic Algorithm II (NSGA II) framework. It features customized encoding/decoding and evolutionary operators to handle multi-device storage allocation, multi-path routing selection and distributed transmission resource allocation. As a result, the best trade-offs among service latency, service failure rate, and MNO cost can be obtained. Simulation results demonstrate that VSR-EA exhibits broader applicability and yields significantly superior solutions compared to benchmark algorithms.

Autonomous Fourier Ptychographic Microscopy (FPM) is a technology widely used in the field of pathology. It is compatible with high resolution and large field-of-view imaging and can observe more image details. Red blood cells play an indispensable role in assessing the oxygen-carrying capacity of the human body and in screening for clinical diagnosis and treatment needs. In this paper, the blood cell data set is constructed based on the FPM system experimental platform. Before training, four enhancement strategies are adopted for the blood cell image data to improve the generalization and robustness of the model. A blood cell detection algorithm based on SCD-YOLOv7 is proposed. Firstly, the C-MP (Convolutional Max Pooling) module and DELAN (Deep Efficient Learning Automotive Network) module are used in the feature extraction network to optimize the feature extraction process and improve the extraction ability of overlapping cell features by considering the characteristics of channels and spatial dimensions. Secondly, through the Sim-Head detection head, the global information of the deep feature map (mean average precision) and the local details of the shallow feature map are fully utilized to improve the performance of the algorithm for small target detection. MAP is a comprehensive indicator for evaluating the performance of object detection algorithms, which measures the accuracy and robustness of a model by calculating the average precision (AP) under different categories or thresholds. Finally, the Focal-EIoU (Focal Extended Intersection over Union) loss function is introduced, which not only improves the convergence speed of the model but also significantly improves the accuracy of blood cell detection. Through quantitative and qualitative analysis of ablation experiments and comparative experimental results, the detection accuracy of the SCD-YOLOv7 algorithm on the blood cell data set reached 92.4%, increased by 7.2%, and the calculation amount was reduced by 14.6 G.

White blood cell (WBC) classification plays a crucial role in hematopathology and clinical diagnostics. However, traditional methods are constrained by limited receptive fields and insufficient utilization of contextual information, which hinders classification performance. To address these limitations, this paper proposes an enhanced WBC classification algorithm, CCE-YOLOv7, which is built upon the YOLOv7 framework. The proposed method introduces four key innovations to enhance detection accuracy and model efficiency: (1) A novel Conv2Former (Convolutional Transformer) backbone was designed to combine the local pattern extraction capability of convolutional neural networks (CNNs) with the global contextual reasoning of transformers, thereby improving the expressiveness of feature representation. (2) The CARAFE (Content-Aware ReAssembly of Features) upsampling operator was adopted to replace conventional interpolation methods, thereby enhancing the spatial resolution and semantic richness of feature maps. (3) An Efficient Multi-scale Attention (EMA) module was introduced to refine multi-scale feature fusion, enabling the model to better focus on spatially relevant features critical for WBC classification. (4) Soft-NMS (Soft Non-Maximum Suppression) was used instead of traditional NMS to better preserve true positives in densely packed or overlapping cell scenarios, thereby reducing false positives and false negatives. Experimental validation was conducted on a WBC image dataset acquired using the Fourier ptychographic microscopy (FPM) system. The proposed CCE-YOLOv7 achieved a detection accuracy of 89.3%, showing a 7.8% improvement over the baseline YOLOv7. Furthermore, CCE-YOLOv7 reduced the number of parameters by 2 million and lowered computational complexity by 5.7 GFLOPs, offering an efficient and lightweight model suitable for real-time clinical applications. To further evaluate model effectiveness, comparative experiments were conducted with YOLOv8 and YOLOv11. CCE-YOLOv7 achieved a 4.1% higher detection accuracy than YOLOv8 while reducing computational cost by 2.4 GFLOPs. Compared with the more advanced YOLOv11, CCE-YOLOv7 maintained competitive accuracy (only 0.6% lower) while using significantly fewer parameters and 4.3 GFLOPs less in computation, highlighting its superior trade-off between accuracy and efficiency. These results demonstrate that CCE-YOLOv7 provides a robust, accurate, and computationally efficient solution for automated WBC classification, with significant clinical applicability.

Autonomous electric vehicle (AEV) services leverage advanced autonomous driving and electric vehicle technologies to provide innovative, driverless transportation solutions. The biggest challenge faced by AEVs is the limited number of charging stations and long charging times. A critical challenge is maximizing passenger travel satisfaction while reducing the AEV idle time. This involves coordinating passenger transport and charging tasks via leveraging the information from charging stations, passenger transport, and AEV data. There are four important contributions in this paper. Firstly, we introduce an integrated scheduling model that considers both passenger transport and charging tasks. Secondly, we propose a multi-level differentiated charging threshold strategy, which dynamically adjusts the charging threshold based on both AEV battery levels and the availability of charging stations, reducing competition among vehicles and minimizing waiting times. Thirdly, we develop a rapid strategy to optimize the selection of charging stations by combining geographic and deviation distance. Fourthly, we design a new evolutionary algorithm to solve the proposed model, in which a buffer space is introduced to promote diversity within the population. Finally, experimental results show that compared to the existing state-of-the-art scheduling algorithms, the proposed algorithm shortens the running time of scheduling algorithms by 6.72% and reduces the idle driving time of AEVs by 6.53%, which proves the effectiveness and efficiency of the proposed model and algorithm.

Network slicing allocation is a new kind of task and resource scheduling technique for 5G and 6G networks and provides the high quality service by allocating the limited infrastructure network resources to massive heterogeneous user requests (various network slices). Currently, most existing studies only consider the situation that infrastructure network resources can meet the demands of all users. However, with the quick increase of network users, the infrastructure network resources usually cannot satisfy the demands of all users. To tackle this problem, this paper designs a two-objective optimization model to maximize the service provider revenue and the user experience. To solve the model effectively, a new evolutionary algorithm based on NSGA-II framework is proposed, in which the tailor-made encoding scheme, decoding scheme and evolutionary operators are designed, respectively. The simulation results indicate that the proposed algorithm can provide diverse resource allocation solutions and is superior in scenarios where the user resource demands exceed the total available resources in the infrastructure network.

The multicast routing problem in software-defined networking (SDN) is an NP-hard problem. The existing solution methods based on deep strength learning suffer from the problems of branch redundancy, an excessively large action space and slow convergence of the intelligent models. In this paper, an intelligent multicast routing algorithm based on deep hierarchical reinforcement learning is proposed to circumvent the aforementioned problems. First, the optimal multicast tree problem is decomposed into two subproblems: fork node selection and the construction of an optimal path from a fork node to a destination node. Second, a multichannel matrix is designed as the state space for the internal and external controllers of hierarchical reinforcement learning based on the global network-aware information characteristics of SDN. Then, different action spaces are designed for the upper and lower subproblems, four action selection policies are designed for constructing multicast paths, and different reward policies are designed at different levels. Finally, a series of experiments and their results show that the designed algorithm not only searches the multicast tree efficiently but also converges faster and without redundant branches, with better performance in terms of bandwidth, delay and packet loss rate than the current mainstream solution algorithms. The codes for DHRL-FNMR are open and available at https://github.com/GuetYe/DHRL-FNMR.

Message transmission and message synchronization for multicontroller interdomain routing in software-defined networking (SDN) have long adaptation times and slow convergence speeds, coupled with the shortcomings of traditional interdomain routing methods, such as cumbersome configuration and inflexible acquisition of network state information. These drawbacks make it difficult to obtain a global state information of the network, and the optimal routing decision cannot be made in real time, affecting network performance. This paper proposes a cross-domain intelligent SDN routing method based on a proposed multiagent deep reinforcement learning. First, the network is divided into multiple subdomains managed by multiple local controllers, and the state information of each subdomain is flexibly obtained by the designed SDN multithreaded network measurement mechanism. Then, a cooperative communication module is designed to realize message transmission and message synchronization between root and local controllers, and socket technology is used to ensure the reliability and stability of message transmission between multiple controllers to realize the real-time acquisition of global network state information. Finally, after the optimal intradomain and interdomain routing paths are adaptively generated by the agents in the root and local controllers, a prediction mechanism for the network traffic state is designed to improve the awareness of the cross-domain intelligent routing method and enable the generation of the optimal routing paths in the global network in real time. Experimental results show that the proposed cross-domain intelligent routing method can significantly improve the network throughput, reduce the network delay and the packet loss rate compared to the Dijkstra and OSPF routing methods.

Fourier ptychographic microscopy (FPM) is a computational imaging technology that has endless vitality and application potential in digital pathology. Colored pathological image analysis is the foundation of clinical diagnosis, basic research, and most biomedical problems. However, the current colorful FPM reconstruction methods are time-inefficient, resulting in poor image quality due to optical interference and reconstruction errors. This paper combines coloring and FPM to propose a self-supervised generative adversarial network (GAN) for FPM color reconstruction. We design a generator based on the efficient channel residual (ECR) block to adaptively obtain efficient cross-channel interaction information in a lightweight manner, and we introduce content-consistency loss to learn the high-frequency information of the image and improve the image quality of the staining. Furthermore, the effectiveness of our proposed method is demonstrated through objective indicators and visual evaluations.

The cross-domain multicast routing problem in a software-defined wireless network with multiple controllers is a classic NP-hard optimization problem. As the network size increases, designing and implementing cross-domain multicast routing paths in the network requires not only designing efficient solution algorithms to obtain the optimal cross-domain multicast tree but also ensuring the timely and flexible acquisition and maintenance of global network state information. However, existing solutions have a limited ability to sense the network traffic state, affecting the quality of service of multicast services. In addition, these methods have difficulty adapting to the highly dynamically changing network states and have slow convergence speeds. To this end, this paper aims to design and implement a multiagent deep reinforcement learning based cross-domain multicast routing method for SDWN with multicontroller domains. First, a multicontroller communication mechanism and a multicast group management module are designed to transfer and synchronize network information between different control domains of the SDWN, thus effectively managing the joining and classification of members in the cross-domain multicast group. Second, a theoretical analysis and proof show that the optimal cross-domain multicast tree includes an interdomain multicast tree and an intradomain multicast tree. An agent is established for each controller, and a cooperation mechanism between multiple agents is designed to effectively optimize cross-domain multicast routing and ensure consistency and validity in the representation of network state information for cross-domain multicast routing decisions. Third, a multiagent reinforcement learning-based method that combines online and offline training is designed to reduce the dependence on the real-time environment and increase the convergence speed of multiple agents.

Mini-batch partitioning is a widely used technique in deep learning that involves dividing a dataset into smaller subsets. This method is crucial in training deep learning models such as deep neural networks and convolutional neural networks. It is favored for its ability to accelerate model convergence, reduce memory overhead, and minimize convergence errors. The primary advantage of mini-batch partitioning is that it allows the model to learn dataset features more evenly, thereby speeding up the convergence process. However, determining the optimal method and size for mini-batch partitioning remains a challenging problem. This paper proposes a novel mini-batch partitioning method focused on feature reorganization. By leveraging a Self-Organizing Map network for feature extraction, data with similar characteristics are initially grouped into the same batch. The purity index of each batch is then calculated based on the number of features and labels, allowing for a comprehensive evaluation of batch homogeneity. Batches with significant differences in purity are selectively reorganized to ensure that each batch contains a diverse set of features, reducing intra-batch feature correlation and ultimately enhancing data representation.Furthermore, through SOM network mapping, the dataset can be effectively partitioned into subsets that are well-suited for model training. Experimental comparisons of various batch partitioning methods on multiple UCI datasets demonstrate that our proposed method, termed MPBS (Mini-Batch Partitioning Algorithm based on Self-Organizing Map Network). Compared with other algorithms, the accuracy, loss and training time are improved by 14.06%, 24.31% and 31.22%.

Multicasting, as a main transmission mode for most of the current applications in elastic optical network, has attracted more and more research attention. In this paper, we investigate multicast scheduling model and solving algorithm. First, we model the multicast scheduling problem as a multi-objective optimization problem (MOP) by minimizing the bandwidth resources and maximizing the user service quality, and then, we propose a deep reinforcement learning assisted multi-objective algorithm for the model (DRL-MM), in which we design source node selection strategy, routing scheme, modulation scheme and spectrum assignment scheme for each multicast session. To identify the superiority of the proposed DRL-MM, we conduct the experiments and compare DRL-MM with an approximation based Steiner tree algorithm (STA-RSA) and a load-balancing routing tree-based algorithm (LD-RSA) through the experiments. The results show that DRL-MM outperforms STA-RSA and LD-RSA in terms of both bandwidth resource usage and user service quality.

Fourier ptychographic microscopy, as a computational imaging method, has the ability to reconstruct high-resolution image, but it suffers optical aberration, which affects its imaging quality. For this reason, this paper proposes a network model to simulate forward imaging process in the Tensorflow framework, and introduces the Wirtinger flow algorithm with samples and coherent transfer function as inputs. Additionally, this paper also optimises the update by back-propagation and accurately estimates aberration using the Zernike polynomials. The superiority of the method is demonstrated through simulations and experiments, which prove that this methods exhibits higher robustness under complex aberration conditions and is able to reduce the impact of optical aberration on imaging quality.

Most of the available divisible-load scheduling models assume that all servers in networked systems are idle before workloads arrive and that they can remain available online during workload computation. In fact, this assumption is not always valid. Different servers on networked systems may have heterogenous available times. If we ignore the availability constraints when dividing and distributing workloads among servers, some servers may not be able to start processing their assigned load fractions or deliver them on time. In view of this, we propose a new multi-installment scheduling model based on server availability time constraints. To solve this problem, we design an efficient heuristic algorithm consisting of a repair strategy and a local search strategy, by which an optimal load partitioning scheme is derived. The repair strategy guarantees time constraints, while the local search strategy achieves optimality. We evaluate the performance via rigorous simulation experiments and our results show that the proposed algorithm is suitable for solving large-scale scheduling problems employing heterogeneous servers with arbitrary available times. The proposed algorithm is shown to be superior to the existing algorithm in terms of achieving a shorter makespan of workloads.

The big data explosion has sparked a strong demand for high-performance data processing. Meanwhile, the rapid development of networked computing systems, coupled with the growth of Divisible-Load Theory (DLT) as an innovative technology with competent scheduling strategies, provides a practical way of conducting parallel processing with big data. Existing studies in the area of DLT usually consider the scheduling problem with regard to fine-grained divisible workloads. However, numerous big data loads nowadays can only be abstracted as coarse-grained workloads, such as large-scale image classification, context-dependent emotional analysis and so on. In view of this, this paper extends DLT from fine-grained to coarse-grained divisible loads by establishing a new multi-installment scheduling model. With this model, a subtle heuristic algorithm was proposed to find a feasible load partitioning scheme that minimizes the makespan of the entire workload. Simulation results show that the proposed algorithm is superior to the up-to-date multi-installment scheduling strategy in terms of achieving a shorter makespan of workloads when dealing with coarse-grained divisible loads.