Current Issue : October-December Volume : 2026 Issue Number : 4 Articles : 5 Articles
To adapt to the complex and volatile economic environment, the significance of a company’s net profit growth forecast for investors’ decision-making, corporate strategic planning, and market regulation has become increasingly prominent. Although the existing prediction methods have made certain progress, their prediction performance is still restricted due to the difficulty in fully capturing the temporal correlation of complex financial data and the low efficiency of model parameter optimization. In view of this, this paper proposes the PSO-Transformer classification algorithm. By integrating the advantages of the Particle Swarm Optimization (PSO) algorithm and the Transformer model to break through the bottleneck, data statistical analysis and correlation analysis are carried out first. And Decision tree, Random Forest, AdaBoost, GBDT, ExtaTrees, CatBoost and XGBoost were selected as the comparison models. From the perspective of evaluation indicators, the Our model has significant advantages in the core dimensions of error control and goodness of fit: its MSE is 29.098 and RMSE is 5.394, both being the lowest among all models. It can optimally control the square term and root mean square error of the prediction error, effectively reducing the interference of large errors on the overall prediction results. The R2 reached 0.432, the highest among all models, significantly higher than AdaBoost (0.369), which ranked second. It has the best ability to explain data variation and the best fit, and can more accurately capture the inherent patterns of the data. The MAE is 4.482, which is slightly higher than ExtaTrees (4.355) and AdaBoost (4.379), but lower than most models such as Random Forest (4.529), CatBoost (4.418), and XGBoost (4.805). The average absolute error control is at an excellent level. Overall, this model performs outstandingly in reducing squared errors and improving the goodness of fit. Only the control of relative errors needs further optimization. It provides an effective new path for improving the accuracy of the company’s net profit growth prediction and has significant practical significance for optimizing investment decisions, assisting enterprise strategic planning, and enhancing the scientific nature of market supervision....
The Triangle Scheduling (TS) Problem, defined by Dürr et al. (J Sched 21:305–312, 2018. https://doi.org/10.1007/s10951-017-0533-1), is a geometric model for nonpreemptive scheduling of jobs with different criticality levels on a single machine. The jobs have a criticality equal to the worst-case execution time and are scheduled off-line. In this article, we describe, implement and analyze the Bintree algorithm on TS, which is an algorithm based on a binary tree construction. It has O(n log(n)) runtime and its approximation ratio is between 1.35 and 2 ln(2) ≈ 1.386. Bintree is, therefore, the first polynomial-time approximation algorithm for TS with an approximation ratio below 1.5. We also explore Bintree’s relation to a previously defined algorithm, Greedy, and a potential hybrid algorithm that runs both and chooses the shorter schedule, which we suspect to be better than either algorithm by itself. We analyze the behavior of Bintree on small values of input sizes formalizing its quadratic integer programming model....
This paper verifies the feasibility of implementing swarm intelligence algorithms based on memristors. Taking the Firefly Algorithm (FA) as a representative swarm intelligence method, we binarize it and integrate it with an orthogonal crossing strategy to enhance its global search capability, thus proposing an orthogonal crossingenhanced memristive binary firefly algorithm (MEM-OC-BFA). This algorithm is deployed on a memristor crossbar array to support image threshold segmentation and feature selection tasks. Experimental results show that the proposed MEM-OC-BFA significantly improves the computational efficiency of image segmentation: in fundus image segmentation tasks, its running time is 1.3-2.7 times shorter than that of the standard BFA and 2.8-6.4 times shorter than that of the brute-force search method, while ensuring no loss of segmentation accuracy with an error tolerance of 0.1%. In feature selection tasks across multiple public datasets (e.g., Leukemia, Musk, Sonar), MEM-OC-BFA exhibits significantly superior average accuracy and stability compared to Particle Swarm Optimization (PSO) and memristive binary firefly algorithm with cosine similarity (MEM-CS-BFA)....
As urbanization picks up pace and the public demand for security keeps climbing, video surveillance systems have emerged as a vital tool for maintaining social stability and safeguarding public safety. Person Re-Identification (Re-ID), as one of the core technologies in intelligent monitoring, mainly aims to accurately match pedestrian identities across cameras without overlapping fields of view. However, in practical applications, occlusion remains a primary challenge that severely degrades Re-ID performance. Especially in high-density crowds, pedestrians are often partially or completely obscured by other objects or individuals, resulting in incomplete image information and impaired feature representation, which significantly reduces recognition accuracy and reliability. Aiming at the problems of excessive reliance on external pose estimation models and asymmetric information matching in occluded Re-ID, this paper proposes a transformer-based pedestrian background decoupling network. The algorithm achieves foreground–background separation and multi-scale feature matching through the synergy of three modules. Meanwhile, a two-stage training strategy is adopted: the first stage optimizes the decoupling module to ensure clean feature separation, while the second stage jointly fine-tunes the correlation module to enhance matching accuracy. Extensive experimental results show that the proposed algorithm outperforms existing methods....
Biological systems, such as cellular metabolism, involve thousands of reactions that together determine how cells grow, respond to their environment, and produce energy. Modeling and analyzing these systems require solving very large mathematical problems that can quickly become computationally prohibitive. To address this challenge, we present a quantum algorithm to analyze metabolic networks, focusing on flux balance analysis as a representative case. We use a quantum interior point method consisting of a quantum subroutine for matrix inversion. Specifically, we reformulate the metabolic optimization problem for efficient execution on a quantum computer using quantum singular value transformation, enabling a rapid solution of complex systems that arise in flux balance analysis. This quantum approach offers a potential computational advantage over classical interior point methods for large and well-conditioned networks. We demonstrate the practical applicability of our method with numerical simulations on the glycolysis and tricarboxylic acid (TCA) cycle network and show that the quantum solution converges to the correct biological objective (objective error < 10−3 and feasibility error < 10−5). This work represents the first application of quantum algorithms to metabolic pathway analysis, establishing a new direction for quantum computational biology and paving the way for quantum approaches to large-scale biological optimization....
Loading....