Although Transformers excel in natural language processing, their extension to time series forecasting remains challenging due to insufficient consideration of the differences between textual and temporal modalities. In this paper, we develop a novel Transformer architecture designed for time series data, aiming to maximize its representational capacity. We identify two key but often overlooked characteristics of time series: (1) unidirectional influence from the past to the future, and (2) the phenomenon of decaying influence over time. These characteristics are introduced to enhance the attention mechanism of Transformers. We propose TimeFormer, whose core innovation is a self-attention mechanism with two modulation terms (MoSA), designed to capture these temporal priors of time series under the constraints of the Hawkes process and causal masking. Additionally, TimeFormer introduces a framework based on multi-scale and subsequence analysis to capture semantic dependencies at different temporal scales, enriching the temporal dependencies. Extensive experiments conducted on multiple real-world datasets show that TimeFormer significantly outperforms state-of-the-art methods, achieving up to a 7.45% reduction in MSE compared to the best baseline and setting new benchmarks on 94.04% of evaluation metrics. Moreover, we demonstrate that the MoSA mechanism can be broadly applied to enhance the performance of other Transformer-based models.

Currently, most spiking neural networks (SNNs) still mimic the chain-like hierarchical architecture in traditional artificial neural networks (ANNs). This method significantly differs from random connections between neurons found in biological brains, limiting the ability to model the evolving mechanisms of neural pathways in biological neural systems, particularly in terms of dynamic depth-scalability and adaptive path-plasticity. This paper develops a new modeling paradigm for SNNs with random graph architecture (RGA), termed Cognition-aware SNN (CogniSNN). Furthermore, we model the depth-scalability and path-plasticity in CogniSNN by introducing a modified spiking residual neural node (ResNode) to counteract network degradation in deeper graph pathways, as well as a critical path-based algorithm that enables CogniSNN to perform path reusability on new tasks leveraging the features of the data and the RGA learned in old tasks. Experiments show that the performance of CogniSNN with redesigned ResNode is comparable, even superior, to current state-of-the-art SNNs on neuromorphic datasets. The critical path-based approach effectively achieves path reuse capability while maintaining expected performance in learning new tasks that are similar to or distinct from the old ones. This study showcases the potential of RGA-based SNNs and paves a new path for modeling the fusion of computational neuroscience and deep intelligent agents. The code is available at github.com/Yongsheng124/CogniSNN.

Existing 3D anomaly detection methods mainly include reconstruction-based methods and memory-based methods. However, reconstruction-based methods rely on anomaly simulation strategies, while the memory bank of memory-based methods cannot cover the features of all points. Different from existing methods, this paper proposes Template3D-AD, a 3D anomaly detection method based on template matching. Template3D-AD matches the test sample with the template based on center points, and extracts the global features and local features of the center point respectively. Considering that the appearance of anomalies is related to the change of surface shape, this paper proposes a curvature-based local feature representation method, which increases the feature difference between abnormal surfaces and normal surfaces. Then, this paper designs a global-local detection strategy, which combines global feature differences and local feature differences for anomaly detection. Extensive experiments show that Template3D-AD outperforms the state-of-the-art methods, achieving 84.4% (1.5% ↑) I-AUROC on the Real3D-AD dataset and 86.5% (11.6% ↑) I-AUROC on the Anomaly-ShapeNet dataset. Code at https://github.com/CaedmonLY/Template3D-AD.

Real-world time series typically exhibit complex temporal variations, making the time series classification task notably challenging. Recent advancements have demonstrated the potential of multi-scale analysis approaches, which provide an effective solution for capturing these complex temporal patterns. However, existing multi-scale analysis-based time series prediction methods fail to eliminate redundant scale-shared features across multi-scale time series, resulting in the model over- or under-focusing on scale-shared features. To address this issue, we propose a novel end-to-end Disentangled Multi-Scale framework for Time Series classification (DisMS-TS). The core idea of DisMS-TS is to eliminate redundant shared features in multi-scale time series, thereby improving prediction performance. Specifically, we propose a temporal disentanglement module to capture scale-shared and scale-specific temporal representations, respectively. Subsequently, to effectively learn both scale-shared and scale-specific temporal representations, we introduce two regularization terms that ensure the consistency of scale-shared representations and the disparity of scale-specific representations across all temporal scales. Extensive experiments conducted on multiple datasets validate the superiority of DisMS-TS over its competitive baselines, with the accuracy improvement up to 9.71%.

Diabetic retinopathy (DR) is a major cause of blindness, but current classification models suffer from low interpretability and difficulty in adjustment. To address these issues, a two-stage deep learning method for DR classification has been proposed, featuring lesion-sliced detection and DR classification stages. In the lesion-sliced detection stage, an improved neural process model extracts information from fundus images by fusing lesion details from various image locations, significantly enhancing accuracy. In the DR classification stage, an enhanced deep forest model was used to identify critical features influencing DR grades, boosting the credibility of the grading outcomes. Tests on the IDRiD and E-ophtha datasets demonstrated superior performance and generalisation ability of the lesion-sliced detection model compared to mainstream neural networks. Meanwhile, experiments on the Kaggle dataset confirmed that the deep forest-based DR classification model outperformed both traditional forest models and residual networks, marking its first application in DR classification. This approach achieves high accuracy and reliability, with improvements in both detection efficiency and generalisation.

As a component of the capital market, the stock market plays a very important role in economic development. How to predict the stock price more accurately is the key concern of researchers and investors. To solve the above problems, this paper proposes an Informer based method for stock intraday price prediction. Informer based on attention mechanism can efficiently capture precise long-range dependencies between input and output to increase the prediction capacity. In this research, historical K-line data of traditional securities in China’s financial market are collected. Based on this data set, the intraday stock price prediction performance of Informer and Long Short-Term Memory neural network (LSTM) is compared. The experiment also uses the three-day K-line model to perform pattern recognition on the K-line data of candidate stocks. The results show that the prediction performance of Informer is significantly higher than that of LSTM, and the data information after three-day K-line pattern recognition can improve the prediction accuracy of Informer and reduce time and space overhead of Informer.

With the rapid expansion of the Internet of Things (IoT), the shift from cloud computing to Mobile Edge Computing (MEC) has become necessary to address the low-latency requirements of real-time applications. Verifiable computation (VC) enables resource-limited clients to outsource their computation-intensive tasks to a powerful cloud while ensuring the correctness of the computation result. However, traditional VC schemes, originally designed for cloud computing, face challenges when applied to MEC environments, such as scalability issues, robustness, and efficiency concerns. To this end, we propose a verifiable distributed computation scheme for MEC, where computation tasks are distributed between a cloud server cluster (consisting of$n$servers) and an edge server. The cloud handles most of the computation through parallel sub-tasks, while the edge server verifies intermediate results and performs minimal computation to recover the final outcome. Our scheme guarantees that the result can be recovered if at least$t$servers, out of a total of$n$servers in the cloud server cluster, perform their computations honestly. By leveraging batch verification and matrix-optimized polynomial evaluations, our scheme significantly enhances scalability, fault tolerance, and efficiency. The extensive analysis and simulations demonstrate that our proposed scheme is more feasible than existing solutions.

Deep ensemble models have been demonstrated to show promising generalization capability. A deep ensemble model includes several deep neural networks as base-learners. Building a deep ensemble model is a challenging task, since maintaining the prediction performance of each base-learner and the diversity among base-learners at the same time is difficult. To address this problem, this paper proposes a two-stage optimization algorithm for deep ensemble model generation, called ELDE-TS. ELDE-TS aims to build a weighted voting-based deep ensemble model for classification tasks end-to-end. The ensemble model includes several convolutional neural network classifiers with different hyperparameters. Each classifier is assigned a weight. The first stage of ELDE-TS is a bi-objective algorithm that generates candidate classifiers for the ensemble model. It takes the validation accuracy and the diversity among classifiers as the optimization objectives. A novel objective function is proposed for the first stage to describe the diversity among the classifiers. The second stage is a single-objective algorithm, which selects representative classifiers for the ensemble model and calculates a weight for each classifier. A tree-based non-repetitive evaluation mechanism is embedded in the second stage to accelerate the search process. The experimental results show that the ensemble model generated by ELDE-TS has competitive performance over the state-of-the-art ensemble models and hand-designed deep models on the Fashion-MNIST, CIFAR-10, and CIFAR-100 datasets. Furthermore, further analysis demonstrates that the proposed ensemble selection method and the non-repetitive evaluation mechanism positively contribute to improving the performance of the ensemble model.

The object detection task is to locate and classify objects in an image. The current state-of-the-art high-accuracy object detection algorithms rely on complex networks and high computational cost. These algorithms have high requirements on the memory resource and computing capability of the deployed device, and are difficult to apply to mobile and embedded devices. Through the depthwise separable convolution and multiple efficient network structures, this paper designs a lightweight backbone network and two different multiscale feature fusion structures, and proposes a lightweight one-stage object detection algorithm—MiniYOLO. With the model size of only 4.2 MB, MiniYOLO still maintains a high detection accuracy, realizing the trade-off between the model size and detection accuracy. Experimental results on MS COCO 2017 data set show that compared to the state-of-the-art PP-YOLO-tiny, MiniYOLO achieves higher mAP with the same model size. Compared with other lightweight object detection algorithms, MiniYOLO has certain advantages in detection accuracy or model size. The code associated with this paper can be downloaded from https://github.com/CaedmonLY/MiniYOLO/.

User association has emerged as a distributed resource allocation problem in the heterogeneous networks (HetNets). Although an approximate solution is obtainable using the approaches like combinatorial optimization and game theory-based schemes, these techniques can be easily trapped in local optima. Furthermore, the lack of exploring the relation between the quality of the solution and the parameters in the HetNet [e.g., the number of users and base stations (BSs)], at what levels, impairs the practicability of deploying these approaches in a real world environment. To address these issues, this paper investigates how to model the problem as a distributed constraint optimization problem (DCOP) from the point of the view of the multiagent system. More specifically, we develop two models named each connection as variable (ECAV) and each BS and user as variable (EBUAV). Hereinafter, we propose a DCOP solver which not only sets up the model in a distributed way but also enables us to efficiently obtain the solution by means of a complete DCOP algorithm based on distributed message-passing. Naturally, both theoretical analysis and simulation show that different qualitative solutions can be obtained in terms of an introduced parameter which has a close relation with the parameters in the HetNet. It is also apparent that there is 6% improvement on the throughput by the DCOP solver comparing with other counterparts when . Particularly, it demonstrates up to 18% increase in the ability to make BSs service more users when the number of users is above 200 while the available resource blocks (RBs) are limited. In addition, it appears that the distribution of RBs allocated to users by BSs is better with the variation of the volume of RBs at the macro BS.

Because of load fluctuating, the performance of service-based system(SBS) in the cloud environment may deviate from service level agreement(SLA). In the cloud environment, it is important to dynamically allocate resource for SBS according to the predicted system load, so as to satisfy global SLA constraint and minimize resource cost. By the analysis of complex business logic in SBS and the feature of dynamic resource allocation problem, this paper models the dynamic resource allocation problem as composite optimization problem and proposes the cost optimization oriented dynamic resource allocation model. Then this paper applies genetic algorithm to solve the dynamic resource allocation model so as to improve the resolving efficiency. Finally, the approach proposed in this paper is evaluated and compared with some related algorithms. It reveals very encouraging results in terms of the quality of resource allocation.

Multiobjective density driven evolutionary algorithm (MODdEA) has been quite successful in solving multiobjective optimization problems (MOPs). To further improve its performance and address its deficiencies, this paper proposes a hybrid evolutionary algorithm based on dimensional diversity (DD) and firework explosion (FE). DD is defined to reflect the diversity degree of population dimension. Based on DD, a selection scheme is designed to balance diversity and convergence. A hybrid variation based on FE and genetic operator is designed to facilitate diversity of population. The proposed algorithm is tested on 14 tests problems with diverse characteristics and compared with three state-of-the-art designs. Experimental results show that the proposed design is better or at par with the chosen state-of-the-art algorithms for multiobjective optimization.

In cloud computing environments, each service-oriented application is often instantiated with multiple workflow instances, and each of them provides a specific QoS level for a particular user category. In recent researches, most of works deploy service instances for each workflow independently, and the sharing of service instance is not considered across compositions related with different SLAs. This may cause several high-performance service instances are exclusively accessed by specific workflow instances, which can incur the following two issues: First, the remaining service instances cannot satisfy SLA for other users when high-performance ones are monopolized by some requests. Second, the utilization ratio of high-performance services will be decreased. To address these problems, a service composition model supporting the instances sharing in multiple SLA environments is provided in this paper. Since the QoS-aware service composition problem is known as NP-hard, which needs a significant amount of computation time to discover optimal solutions, a solving method based on multi-objective genetic algorithm, called MSCS, is proposed to solve it heuristically. The effectiveness of this approach is demonstrated via experiments.

Reselection of composition service is one of the core research issues in the service computing field. Most of the existing researches for this problem are based on the assumption that the tasks involved are independent. However, in practical scenar-ios, the QoS of some candidate services have correlations with other services, which makes the corresponding tasks be correlated with each other. This leads the QoS used to determine the binding relationship between tasks and concrete services to be inaccurate, so the reselected composite service is not the optimal one in the real executing environment for these existing reselection methods. To address this problem, this paper considers task correlations for runtime rebinding. Firstly, the QoS dependencies among services are extracted from the log repository through the APRIORI data mining method. Then, the acquired QoS dependencies are mapped to the tasks correlations by the defined mapping rules. Finally, the reselection process is implemented by making the tasks which have related relationships as a task unit, and the related services of each task unit as its candidate service set. The effectiveness of this approach, in terms of time and quality, is demonstrated via experiments.