Designing a data center network (DCN) architecture capable of accommodating and managing thousands of servers is crucial for optimizing computing services in fields, such as Big Data, cloud computing, and artificial intelligence. This article introduces a novel class of DCN architecture, termed hypercube-like data center network (HLDN), which extends existing frameworks like high scalability data center network and crossed cube-based scalability data center network, while exploring previously unexamined designs. As the scale of DCNs grows and the number of servers increases, the incidence of network failures also rises, impacting network reliability. To address this challenge, we propose the first adaptive diagnostic schemes specifically designed for DCNs. We investigate the Hamiltonian properties of HLDN and, based on this analysis, present parallel adaptive diagnostic algorithms under the Preparata, Metze, and Chien (PMC) and comparison (COM) models. Simulation experiments validate the effectiveness of these algorithms, showing that the PMC model-based approach significantly outperforms the COM model in terms of diagnostic performance under identical conditions. Furthermore, comparative analyses with existing methods highlight the enhanced diagnostic accuracy of our proposed algorithms. This work not only provides an innovative solution for improving the reliability of DCNs, but also offers valuable theoretical and practical insights for fault detection and management in large-scale network systems.

We establish a global adaptive group maintenance policy for multicomponent systems over an infinite time horizon. This process enables the dynamic integration of 1) scheduled group maintenance and 2) unscheduled opportunistic maintenance in a global adaptive manner. Unlike traditional maintenance models that rely on predefined maintenance windows and schedules, our approach continuously adjusts based on the latest operational data, obviating the need to prespecify a scheduling horizon. The group maintenance model is governed by a two-phase scheduling framework: from tentative component age determination to dynamic system group clustering. We provide structural insights into the optimality of system maintenance timing through penalty analysis. Leveraging these findings, we design an efficient heuristic grouping algorithm that transforms the long-term maintenance planning problem into short-term group selection problems, significantly streamlining computational efforts and allowing for adaptive refinement. Comparative experiments on phased-array radar systems demonstrate substantial improvements over existing policies.

The effectiveness and robustness of medical image segmentation algorithms inextricably depends on the quality and quantity of training data. Insufficient data continually restricts model training and affects segmentation accuracy and robustness. We rigorously create a reliable image segmentation model to tackle the challenges presented by constrained medical image datasets. Our method integrates two novel components: the coordinate attention and identity (CAI) module and the Kolmogorov–Arnold and squeeze-and-excitation (KASE) module. The CAI module consists of two branches that enhance feature representation. The initial branch utilizes coordinate attention techniques to detect cross-channel dependency. This offers location-specific and directionally aware information crucial for precisely identifying target areas. The second branch utilizes features produced by convolution as identity mappings. This tackles problems such as vanishing and exploding gradients while maintaining essential features and semantic information. The KASE module comprises two branches. The initial branch improves weight calculations by extracting global context through multiscale squeeze-and-excitation methods. The other branch integrates Kolmogorov–Arnold networks with convolutional operations. This enhances the capacity of the proposed model to learn and represent complex features. These two modules enhance the robustness of the model and improves its generalization across varied input data. Finally, we train multiple state-of-the-art models on multiple datasets using 100%, 20%, 10%, and 5% of the training set in each dataset. The test results show that our model achieves competitive performance despite limited training data.

Ensuring the safety of autonomous driving systems (ADSs) through rigorous verification in simulated environments is crucial before real-world deployment. However, using simulation environments for ADS testing and verification poses several challenges, including specifying various behaviors of traffic participants and collecting comprehensive real-time data to verify the ADS. To address these challenges, we propose a framework for runtime verification of ADSs, focusing on Autoware, a leading ADS. This framework integrates AWSIM-Script, a scripting language for defining traffic scenarios; Runtime Monitor, a tool to record real-time data during the simulation; and AW-Checker, a Linear Temporal Logic-based property checker for verifying safety requirements. Unlike prior research that primarily focuses on generating critical scenarios, we leverage a well-established ADS safety standard from the Japan Automobile Manufacturers Association and adopt its systematic methodology for safety assessment. We conducted a series of experiments focused on nonintersection road geometry to evaluate Autoware's capability in handling different traffic disturbances such as cut-in, cut-out, and deceleration scenarios. The results revealed that, compared to the competent and careful driver model, which represents the minimum safety requirements for ADSs, Autoware failed to prevent collisions in some cases, particularly during high-speed scenarios and fast lateral movements by other vehicles.

Network clustering (graph clustering) plays a crucial role in discovering the inherent structures within networks. MapReduce-based structural clustering algorithm for networks (MRSCANs) designed based on MapReduce's parallel computing model, efficiently handles large-scale data. However, MRSCAN can only be tested through experiments, and its correctness cannot be guaranteed. To address this issue, this article has achieved the first implementation of functional MRSCAN modeling and subjected it to rigorous mechanized verification in Isabelle. First, based on Google's MapReduce model type definition and higher-order generic functions, a general MapReduce-based algorithm functional modeling framework is constructed across the fundamental global phases of Map, Shuffle, and Reduce. Moreover, diverse strategies are devised during the Shuffle phases according to user requirements, enhancing the applicability and generality of the MapReduce functional modeling framework. Second, formalizing the definition of MRSCAN, is delineated into four key steps: similarity calculation, core calculation, dimension expansion, and structural clustering. Furthermore, the MapReduce functional modeling framework is applied to these four steps to achieve the functional modeling of MRSCAN, which improves efficiency compared to other structural clustering algorithm for networks (SCANs) algorithms. Lastly, a verification framework for MapReduce-based algorithms is proposed at both the global and shuffle stages. Based on this framework, the correctness and reliability of MRSCAN are ensured. The model framework and verification framework of MapReduce-based algorithms proposed in this article can not only address functional modeling and verification of MRSCAN but also provide a reference for a series of other MapReduce-based functional program designs and proofs.

With the extensive application of artificial intelligence (AI) technologies, automated programming language-natural language (PL-NL) systems have gained significant attention, driving a series of related tasks served for developers and users, such as code search and summarization. Currently, mainstream PL-NL systems regard PL-NL as bimodal data and utilize two individual neural architectures (e.g., recurrent neural network) to learn the representation of PL-NL and build their semantic relations, improving the effects of these tasks. However, there exist two issues that limit the ability of these service systems in representation learning: first, large vocabularies cause data sparsity problems and limit the learning ability of neural architectures; second, there is not always a one-to-one correspondence between source code and natural language. To address these two issues, in this article, we introduce the unified neural architecture (UNA) by building a unified vocabulary (Uni-Vocab) at the subword level, to provide high-quality PL-NL services. In the Uni-Vocab, we build a unified modal encoding for PL-NL, which allows us to effectively control the vocabulary size and solve the data sparsity problem. Afterward, our built UNA can learn the unified contextual representation of PL-NL, which helps build their unified semantic relations. To validate the effectiveness of the proposed UNA, we perform experiments on code search and code summarization, which are two PL-NL tasks for developers and users. Experimental results demonstrate UNA can obtain noteworthy performance improvement. In detail, the baseline approaches in these two tasks get improvements by up to 36.09% and 18.02% in terms of mean reciprocal rank and bilingual evaluation understudy, respectively.

In recent years, the quantification of structural uncertainty using evidence theory has been widely adopted. However, there is a long-term difficult problem of overconservative reliability in this field. This article reveals that the “one-vote veto” phenomenon in the mapping mechanism of the focal element's basic probability assignment (BPA) to the reliable domain is the fundamental reason. Thus, the rational mapping of focal element's BPA to the reliable domain is a new problem that has not yet been paid attention to and is in urgent need of a solution. To address this new scientific problem, a “cut-merge” approach for uncertainty quantification inspired by triangular distribution is proposed. Its core idea is “cutting” first and then “merging.” “Cutting” is to construct two extreme probability density functions (PDFs) that obey the triangular distribution on each focal element, and “merging” is to integrate the extreme joint PDF on the reliable domain, to realize the scientific mapping of focal element's BPA. This method eliminates the need for extreme value analysis of focal elements and requires only the computation of two definite integrals to obtain the reliability interval. The results of an artificial synthetic case with a known true distribution of focal elements show that the proposed method is obviously superior to the traditional method in most engineering scenarios. The results of four application examples show that compared with the traditional method, the proposed method reduces the uncertainty interval width by 64.81% on average and increases the smoothness index by 67.01% on average. This work provides a new theoretical tool for structural reliability analysis.

Software defect prediction technology aids the reliability assurance team in identifying defect-prone code and assists the team in reasonably allocating limited testing resources. Recently, researchers assumed that the topological associations among code fragments could be harnessed to construct defect prediction models. Nevertheless, existing graph-based methods only concentrate on features of single-view association, which fail to fully capture the rich information hidden in the code. In addition, software defects may involve multiple code fragments simultaneously, but traditional binary graph structures are insufficient for representing these multivariate associations. To address these two challenges, this article proposes a multiview hypergraph representation learning approach (MVHR-DP) to amplify the potency of code features in defect prediction. MVHR-DP initiates by creating hypergraph structures for each code view, which are then amalgamated into a comprehensive fusion hypergraph. Following this, a hypergraph neural network is established to extract code features from multiple views and intricate associations, thereby enhancing the comprehensiveness of representation in the modeling data. Empirical study shows that the prediction model utilizing features generated by MVHR-DP exhibits superior area under the curve (AUC), F-measure, and matthews correlation coefficient (MCC) results compared to baseline approaches across within-project, cross-version, and cross-project prediction tasks.

The complexity of Machine Learning (ML) systems increases each year. As these systems are widely utilized, ensuring their reliable operation is becoming a design requirement. Traditional error detection mechanisms introduce circuit or time redundancy that significantly impacts system performance. An alternative is the use of Concurrent Error Detection (CED) schemes that operate in parallel with the system and exploit their properties to detect errors. CED is attractive for large ML systems because it can potentially reduce the cost of error detection. In this paper, we introduce Concurrent Classifier Error Detection (CCED), a scheme to implement CED in ML systems using a concurrent ML classifier to detect errors. CCED identifies a set of check signals in the main ML system and feed them to the concurrent ML classifier that is trained to detect errors. The proposed CCED scheme has been implemented and evaluated on two widely used large-scale ML models: Contrastive Language-Image Pre-training (CLIP) used for image classification and Bidirectional Encoder Representations from Transformers (BERT) used for natural language applications. The results show that more than 95% of the errors are detected when using a simple Random Forest classifier that is orders of magnitude simpler than CLIP or BERT.

Frequent and sizable voltage fluctuation, a common issue faced by the modern distribution system (DS), could lead to potential equipment failures and power interruption. This brings huge negative impact on the power supply reliability of the DS. Existing voltage regulation methods typically rely on the precise physical parameters, complete measurements, and perfect communication, all of these premises are difficult to meet in practice. To this end, a decentralized control method that is robust to measurement acquisition errors is developed for DS in this article. Specifically, a graph learning-based surrogate network is first built to simulate the power flow computing procedure and capture the structural characteristics of the DS. The centralized surrogate model is, then, divided into several decentralized representation networks according to the network partition results to obtain the robust embedding of the regional information of each subnetwork. Subsequently, the representation networks are embedded in the front of the actor networks of the multi-agent soft actor-critic algorithm, the agents of which are learned in a centralized fashion according to the reward value estimated by the centralized surrogate model. The systematic integration of the three components allows us to achieve cooperation between different subregions and robustness against anomalous measurements without the reliance on precise circuit parameters. Comparative studies on IEEE test system illustrate the robustness of the proposed approach.