Applied Soft Computing最新文献

In the contemporary landscape of the web, text-to-image generation stands out as a crucial information service. Recently, deep learning has emerged as the cutting-edge methodology for advancing text-to-image generation systems. However, these models are typically constructed using domain knowledge specific to the application at hand and a very particular data distribution. Consequently, data scientists must be well-versed in the relevant subject. In this research work, we target a new foundation for text-to-image generation systems by introducing a consensus method that facilitates self-adaptation and flexibility to handle different learning tasks and diverse data distributions. This paper presents I2T-SP (Image-to-Text Generation for Shapley Pruning) as a consensus method for general-purpose intelligence without the assistance of a domain expert. The trained model is developed using a general deep-learning approach that investigates the contribution of each model in the training process. Multiple deep learning models are trained for each set of historical data, and the Shapley Value is determined to compute the contribution of each subset of models in the training. Subsequently, the models are pruned according to their contribution to the learning process. We present the evaluation of the generality of I2T-SP using different datasets with varying shapes and complexities. The results reveal the effectiveness of I2T-SP compared to baseline image-to-text generation solutions. This research marks a significant step towards establishing a more adaptable and broadly applicable foundation for image-to-text generation systems.

Accurate forecasting information of short-term subway passenger flow is an important scientific reference for daily operations and urban management. The rapid time-varying nature of subway passenger flow caused by various factors that affect travel behavior poses a huge challenge to accurate forecasting. The complexity and uncertainty of data mainly focus on residual terms that deeply reflect the fluctuations. Reasonably mining the residual terms has become the key to achieving accurate forecasting. Therefore, this paper proposes a sophisticated decomposition strategy to extract and analyze the useful information hidden in the residual terms. Firstly, the potential trend, seasonal and residual terms from the original data are extracted by seasonal-trend decomposition procedure based on loess. Secondly, the residual term is decomposed into a series of subcomponents with different frequency features by symplectic geometric mode decomposition. Thirdly, these subcomponents are classified into three clusters based on fuzzy C-means clustering (FCM), and then the corresponding forecasting models are matched to the three obtained clusters and trend and seasonal terms. Finally, based on a decomposition-ensemble framework and information granulation for high-frequency components, we have established point and interval forecasting approach, respectively. Three experiments on real data sets in Beijing, Guangzhou and Shenzhen are conducted to verify the performance of the proposed approach, and the experimental results show that our approach is superior to all benchmark models and contributes to improving operational management and service quality.

Efficient truck dispatching strategies are paramount in container terminal operations. The quality of these strategies heavily relies on accurate and expedient simulations, which provide a crucial platform for training and evaluating dispatching algorithms. In this study, we introduce data-driven machine learning methods to enhance container port truck dispatching simulation accuracy. These methods effectively surrogate the intersections within the simulation, thereby increasing the accuracy of simulated outcomes without imposing significant computational overhead in sparse data environments. We incorporate three data-driven learning methods: genetic programming (GP), reinforcement learning (RL), and a GP and RL hybrid heuristic (GPRL-H) approach. The GPRL-H method proved the most efficacious through a detailed comparative study, striking an effective balance between simulation accuracy and computational efficiency. It reduced the error rate of simulation from approximately 35% to about 7%, while also halving the simulation time compared to the RL-based method. Our proposed method also does not rely on precise Global Positioning System (GPS) data to simulate truck operations within a port accurately. Demonstrating robustness and adaptability, this approach holds promise for extending beyond port operations to improve the simulation accuracy of vehicle operations in various scenarios characterized by sparse data.

Deep learning-based image inpainting techniques have achieved unprecedented results using encoder–decoder structures to recover complex missing areas of an image. Recent inpainting models use additional information or networks (e.g., landmarks, edges, styles, and filters) to realize improved restoration performance, but at the cost of increased computational resources. To improve the relationship between inpainting performance and the number of model parameters, researchers have investigated efficient structural approaches such as recurrent and residual connection structures. However, these methods are difficult to apply in the general encoder–decoder structure. In this study, we explored the downsampling and upsampling operations associated with an encoder–decoder structure. We propose two novel split operations: the pixel-split operation (PSO) and channel-split operation (CSO). The proposed PSO transfers image features from high to low resolution with two dilation rate effects and a similar number of parameters as existing downsampling operations. Conversely, the proposed CSO increases the image resolution using only one-fourth the number of parameters of existing upsampling operations. The restoration performance and efficiency of the proposed model were evaluated in terms of five metrics on public datasets, e.g., the Places2 and CelebA datasets, to validate our proposed operations’ contribution to inpainting performance. We achieved state-of-the-art performance and reduced the size of the parameters by 20%. An ablation study was conducted to confirm the effect of each operation on the CelebA-HQ dataset. Results indicated that these split operations exhibit an advanced relationship between inpainting performance and optimization of the model parameters. The corresponding codes are available online (https://github.com/MrCAIcode/Split_operation_for_inpainting).

Allergic reactions can range from mild discomfort to life-threatening situations. To manage the healthcare difficulty, an efficient allergens mapping is required by mapping the allergies to reduce the severity risk reactions. The allergies are mapping with specific food items according to the daily usage. It enables the clear communication, targeted avoidance, and public awareness, which is necessary to scrutinize the prevention process. Since, the allergies are highly individualized and can exhibit significant variability among individuals that are capturing the complexity in the process of overcoming the predictive challenges. The complex relationships and data challenges require advanced approaches like ML and graph theory. For this purpose, we propose a new multidisciplinary framework that integrates the Machine Learning (ML), Graph Theory and Game Theory to predict the allergies associated with relevant foods using a modest dataset. This framework has two newly built graph models such as a conventional approach and a refined approach, to pave the way for better results. Here, the ML techniques are employed to perform the classification process on tabular data that are observed the remarkable improvements and transforming the data into a graph. Moreover, the Darwinian decision-making framework is adopted theoretically in evolutionary game theory to formulate effective formulas for assessing the spread of allergies among allergens and predict the allergies dynamically. The proposed framework has been evaluated by conducting experiments by using a modest dataset by considering the evaluation metrics such as accuracy, macro-precision, macro-recall, and macro F1-score.

Emotion recognition and understanding plays a crucial role in various domains, including healthcare, human-computer interaction, and mental well-being. In this context, this paper proposes a methodology for recognizing and curing emotions using acoustic features and machine learning algorithms. The approach involves extracting acoustic features from the signals using diverse signal processing techniques. These features are then utilized as inputs for machine learning and deep learning algorithms, including the Random Forest classifier, XG Boost classifier, Convolutional Neural Network (CNN), and an ensemble algorithm. The ensemble algorithm combines Random Forest and XG Boost as base classifiers, with the Naïve Bayes algorithm serving as the meta classifier. We also propose a novel model that generates personalized curing strategies for individuals based on emotion recognition, so they can keep their emotional state positive. With the help of an ensemble learning model the proposed model achieved an emotion recognition accuracy of 92 % by combining three publicly available datasets containing emotional speech recordings. In the neutral and positive emotion classifications, the Receiver Operating Characteristic curve (ROC) had a 98 % accuracy rate while negative emotion classifications had a 91 % true positive rate. The effectiveness of the proposed curing methodology model has also been demonstrated by conducting experiments on a group of individuals and comparing the results with a state-of-the-art Generative Pre-Trained Transformer-3 (GPT-3) and ChatGPT, it was inferred that 89.35 % of the test group preferred the responses of the proposed curing model, over the GPT models The results of our experiments show that our proposed methodology can significantly boost the emotional state of an individual, thereby highlighting its potential for use in clinical settings.

The uncertainty and spatiotemporal dynamics of information necessitate the urgent modeling of fuzzy spatiotemporal knowledge across various applications, with the Resource Description Framework (RDF) serving as a widely recognized data representation model. Existing models suffer from incomplete semantic representation and poor robustness in modeling fuzzy spatiotemporal data, e.g., lack of fuzziness in spatiotemporal semantics; lack of altitude in spatial semantics. Meanwhile, the algebraic framework regarding the model has not been investigated. Thus, in this paper, we first propose a new fine-grained fuzzy spatiotemporal RDF model. This model can represent fine-grained uncertain spatiotemporal semantics that may be associated with any element of a spatiotemporal RDF. We further define its graph algebraic operations. Note that we demonstrate the use of the algebraic operations for fuzzy spatiotemporal RDF querying. Finally, we establish a set of transformation rules for SPARQL query syntax to algebraic operations in fuzzy spatiotemporal RDF. In addition, we used experiments to evaluate the validity and rationality of our model.

This paper studies a real-world electric vehicle routing problem (EVRP). Specifically, it is an EVRP with time windows and battery swapping stations (EVRP_TWBSS). The EVRP_TWBSS considers the routing of electric vehicles (EVs), the determination of each electric vehicle’s battery level, and the selection of battery swapping stations. The criterion of EVRP_TWBSS is to minimize the operating costs. To simplify the structure of model, a time-discrete and multi-commodity flow model based on extended state-space-time network (TMFM_ESSTN) is established. Meanwhile, an alternating direction multiplier method with variable neighborhood search (ADMM_VNS) is presented to address the TMFM_ESSTN. In ADMM_VNS, the augmented lagrangian relaxation (ALR) model constructed from the TMFM_ESSTN is decomposed and linearized to a series of least cost vehicle routing subproblems through the linear augmented lagrangian relaxation (LALR) decomposed technique. Then, each subproblem is iteratively solved by using the dynamic programming and two special designed VNS strategies in ADMM_VNS iterative framework. The solution’s quality can be controlled to a certain extent through monitoring the gap between the lower and upper bounds obtained after each iteration. Test results on instances with different scales and a real-world instance based on partial road network in Kunming City demonstrate that ADMM_VNS can achieve smaller gaps and better solutions than several state-of-the-art algorithms. In which, ADMM_VNS can reduce the optimal gap by up to 2.27 % compared to the other state-of-the-art algorithms in small-scale instances. The gap of ADMM_VNS is calculated based on the lower bound and the upper bound in the large-scale instances and the real-world instance are 10.36 % and 1.57 %, respectively.

A novel approach is proposed for multiband image processing via quantum models in real situations. Quantum circuits are automatically generated ad-hoc for each use case via multiobjective genetic algorithms. Using this universal method, image processing tasks such as segmentation can be carried out by considering the properties that constitute each pixel. The generated circuits present a low level of correlation between qubits, and thus can be considered quantum-inspired machine learning models. The effectiveness of this methodology has been validated by applying it to different segmentation use cases. Comparisons are made between optimized classical kernel methods and the generated quantum-inspired models to understand their behaviors. The results show that quantum models for multiband image processing achieve accuracies similar to those of classical methods.

Accurately detecting the location and intensity of impacts is crucial for ensuring structural safety. Currently, AI-based structural impact detection methods are widely used for their excellent detection accuracy. However, their generalization capability is limited by the scenarios present in the training data. Many complex and dangerous impact scenarios are difficult to conduct real-world experiments on to collect sufficient samples. To capture all impact scenarios and fully leverage the advantages of AI-based detection technologies, advanced methods involve combining real-world structural monitoring data with corresponding numerical models to construct digital twins. These methods continuously refine the created numerical models with limited real-world data and provide diverse impact scenarios through numerical model simulations. However, there are inevitable differences between digital models and physical models that are challenging to correct through mechanical means. This discrepancy in data distribution between the two models significantly hinders the application of digital twin technology in impact/event identification tasks. To address this challenge, this study proposes a novel model based on autoencoders, named Transfer-AE. Transfer-AE encodes the common features of digital twins in the latent space to bridge the uncertainty gap at a macro scale between numerical models and physical models and synchronously fits the magnitude and location of the impact load in the decoder. This enables consistent detection results for the same impact event, whether the sample comes from the numerical model or the physical model. Transfer-AE includes two operating modes: Mode 1 has a fixed computational complexity with stable inference speed, but the training cost and difficulty increase with data distribution. Mode 2's computational complexity increases with data distribution, but it has a fixed training cost and speed. In both cases involving the geodesic dome structure simulating a deep space habitat and the IASC-ASCE benchmark structure, Transfer-AE demonstrated the best performance in impact localization and quantification tasks compared to mainstream domain-adaptive transfer models.