The design of optimizer parameters a˙ects model performance and is widely applied in fields such as image analysis, autonomous driving, and security monitoring. However, the interpretability and generalizability of optimizers are insuÿcient, limiting their practical applications. To address these challenges, we introduce a novel approach using transfer function and phase trajectory methods to design the parameters and critical conditions for Stochastic Gradient Descent with Momentum (SGD-M) and Nesterov Accelerated Gradient (NAG). The proposed theory is verified through numerical examples and image recognition experiments. First, using the phase trajectory method, a qualitative analysis of the responses of SGD-M and NAG to initial states is conducted, revealing the influence of parameters on the phase trajectory. Then, through the transfer function method, a quantitative analysis of the unit step response of SGD-M and NAG is performed to explain the impact of parameters on system response. Finally, numerical examples and image recognition experiments verify the significant impact of the momentum control parameter g(µ) and momentum parameter α on optimizer performance, stability, and time-domain characteristics. Experimental results show that adjusting g(µ) or α improves image classification accuracy on the Modified National Institute of Standards and Technology (MNIST) and Canadian Institute for Advanced Research (CIFAR-10) datasets. It reduces the loss value, validating the e˙ectiveness of the proposed theory.

The development of medical big data analysis technology has enabled vast imaging datasets to play an important role in identifying complex pathologies, supporting individualized diagnoses, and enhancing diagnostic and treatment precision. Pediatric pneumonia, a leading cause of mortality in children, results from infections and is characterized by inflammation of the terminal airways, alveoli, and pulmonary interstitium. Accurate and timely diagnosis is essential for successful intervention. However, most current studies are constrained by simplistic classification tasks, and existing models struggle to effectively extract multi-scale features from Chest X-Ray (CXR) images while eliminating redundant information. Moreover, the datasets used in existing studies lack the diversity and scale necessary to meet the demands of clinical practice. To address these challenges, this paper presents the a multi-branch dual residual attention Convolutional Neural Network (CNN) model for pediatric pneumonia classification model (namely PP-ConvNeXt), designed to enable automatic diagnosis of pediatric pneumonia. First, the model incorporates a novel multi-branch dual-residual structure, enhancing its capacity to extract multi-scale features. Second, the Coordinate Attentional Feature Fusion (CAFF) module is introduced to integrate cross-branch features, allowing for comprehensive acquisition of lesion information. Finally, this research pioneers the automatic five-class diagnosis of pediatric pneumonia, offering clinicians more accurate diagnostic insights. Additionally, a comprehensive dataset comprising 5632 CXR images from Hainan Women and Children’s Medical Center (HWCMC) is presented. Experimental results indicate that the PP-ConvNeXt model surpasses state-of-the-art models and expert human diagnosticians in the five-class diagnosis of pediatric pneumonia, achieving an Area Under the Curve (AUC) score of 89.01%. Additionally, key performance metrics, including accuracy and recall, further validate the model’s superior performance.

Cyber-Physical-Social Intelligence (CPSI) enhances the interpretability of Gradient Descent (GD) optimizers and improves image recognition models. To address the black-box nature of optimizers, we propose a new approach to map the GD optimizer to the frequency domain, allowing analysis of gradient variations at different frequencies. This approach aids in selecting optimal training and optimization strategies, offering a novel solution to the challenge of optimizer interpretability. Specifically, firstly, based on the transfer function of the gradient descent optimizer, the real-frequency characteristic and imaginary-frequency characteristic functions of the optimizer are derived for the first time, which provide a new perspective for the selection of the optimizer and the interpretability of the tuning parameter. Then, based on these characteristic functions, the learning properties of the optimizer in the frequency domain are analyzed for the first time, which provides an important reference for improving the performance of the optimizer in real application problems. Finally, the effectiveness of the frequency domain modulation properties is verified through convex and non-convex optimization problems. Experimental results show that the proposed theory not only improves the recognition accuracy, convergence speed, and stability, but also extends its scope in practical applications.

Traffic accident data analysis and reasoning are crucial for accident prevention and control. Constructing an accident knowledge graph from hybrid datasets of Chinese and English accidents is a valuable task. However, it is challenging due to the need to consider multiple perspectives and infer implicit relationships between actors and factors in complex traffic accidents. To address these challenges, this paper proposes an accident learning attention embeddings based multi-view accident knowledge graph for traffic accident reasoning named BrightAccidentGraph. First, this paper proposes a multi-source traffic accident dataset construction and preprocessing method for traffic accident judgement records published by the China Judgement Document Network and traffic accident records published by the UK’s Ministry of Transport. Then, traffic accident graph construction and portrait method is proposed, we demonstrate the efficiency of the proposed method by constructing several multi-view traffic accident portraits using a multi-source dataset. Furthermore, accident learning attention embeddings based multi-view accident knowledge graph construction and traffic accident reasoning method using deep learning are introduced. Experiments on two hybrid datasets verify the efficiency and merits of our method.

Infrared spectroscopy analysis has found widespread applications in various fields due to advancements in technology and industry convergence. To improve the quality and reliability of infrared spectroscopy signals, deconvolution is a crucial preprocessing step. Inspired by the transformer model, we propose an Auto-correlation Multi-head attention Transformer (AMTrans) for infrared spectrum sequence deconvolution. The auto-correlation attention model improves the scaled dot-product attention in the transformer. It utilizes attention mechanism for feature extraction and implements attention computation using the auto-correlation function. The auto-correlation attention model is used to exploit the inherent sequence nature of spectral data and to effectively recovery spectra by capturing auto-correlation patterns in the sequence. The proposed model is trained using supervised learning and demonstrates promising results in infrared spectroscopic restoration. By comparing the experiments with other deconvolution techniques, the experimental results show that the method has excellent deconvolution performance and can effectively recover the texture details of the infrared spectrum.