Muhammad RajabinasabFarhad PakdamanArthur ZimekMoncef Gabbouj...
126254.1-126254.15页
查看更多>>摘要：k-Nearest Neighbors (kNN) search is the problem of finding k points which are the closest to a given query point. It is used widely in a wide range of tasks and is among the most important tools in applied machine learning. Traditional algorithms for kNN search require computing distances between a query point and all other points in the dataset, and therefore is very slow and inefficient for large data. In this paper, we propose an approximate algorithm for kNN search to find the nearest neighbors fast and efficiently. We employ a tree-based structure which offers robustness and scalability. We propose to use Principal Component Analysis (PCA) to find the best splitting direction to fit the data on the trees. Seeking solutions with low computational complexity, (1) we use a randomized Singular Value Decomposition solver, which reduces PCA complexity from being associated with the number of features to being associated with the number of required principal values; (2) we reuse PCA calculations in multiple nodes to save computation while maintaining accuracy; (3) we ensemble these trees for improved performance, and (4) finally, we propose several variants of the proposed method which target a higher accuracy or a higher efficiency. Extensive experimental results show that proposed solutions outperform existing methods in terms of accuracy, while maintaining competitive complexity. The fast implementation variant of the proposed method outperforms existing techniques in terms of complexity and shows competitive accuracy in performing k-nearest neighbors' search.
Kabeer Ahmed BhattiBilal RaufImran Ali QureshiAwais Majeed...
127254.1-127254.9页
查看更多>>摘要：Congestion is an unavoidable problem in the Internet of Things (IoT) network because it is equipped with non-standardized devices. The many-to-one and carry-to-send nature of nodes leads to congestion. The underlying node becomes a bottleneck and faces serious communication problems due to unbalanced internal network parameters. This research presents an adaptive source transmission rate optimization algorithm to resolve these issues by performing a runtime fine-tuning of the proportional integral derivative (PID) controller by applying Non-Dominated Sorting Genetic Algorithm Ⅲ PID (N3PID) in a cascaded manner. The optimal adjustment of internal network parameters results from overcoming the drawbacks of insufficient diversity, slow convergence, and overshoot. The N3PID provides a more accurate response due to efficient and robust parameter tuning. Moreover, the optimally modified parameters are passed to a PID controller that uses the error (the variation between the instantaneous and predicted queues) as input to optimize the transmission rate for the origin node. The N3PID increases the convergence speed and accelerates the accuracy. The N3PID algorithm is assessed with PID, Particle Swam Optimization-neural PID (PNPID), Cuckoo Fuzzy PID (CFPID), and Neural Network PID (NNPID) through a simulation in Network Simulator software. The experimental results reveal that the packet delivery ratio is increased by 9.924% and the average delay is substantially reduced by 14.152% while packet loss is significantly reduced by 12.311% and minimized the energy consumption to 5.899% as compared with NNPID.
查看更多>>摘要：Crop diseases present a critical threat to global food security, with traditional methods for disease detection often relying on manual diagnosis, which is labor-intensive and requires specialized expertise. While deep learning provides promising avenues for automated disease detection, many existing models are challenged by domain shift, performing poorly when applied to data from new environments with different distributions. To address this issue, we propose a novel cross-domain detection framework that integrates Multi-Domain Data Augmentation and Isolated Test-Time Adaptation Optimization (ITTA) to enhance model generalization and robustness in dynamic agricultural settings. Our approach begins with Multi-Domain Data Augmentation, which combines strong and weak augmenters to generate diverse cross-domain images. The strong augmenter, based on Generative Adversarial Networks (GANs), produces varied images that preserve class consistency, while the weak augmenter applies Gaussian noise and brightness adjustments to make the model more resilient to minor environmental changes. This dual augmentation strategy enables the model to learn domain-invariant features and improves its generalization capability across different domains. The second component ITTA, employs a teacher-student model structure that leverages the Fisher Information Matrix (FIM) to isolate domain-sensitive and domain-invariant parameters. During test-time adaptation, only the domain-sensitive parameters in the student model are updated, allowing the model to adapt to new domain-sensitive features while preserving essential knowledge from the source domain. This selective update process prevents catastrophic forgetting and maintains high detection accuracy under varying environmental conditions. Experimental results demonstrate that our framework significantly enhances model performance across diverse settings, offering a robust foundation for intelligent and precise crop disease detection, which is essential for improving agricultural productivity and food security.
查看更多>>摘要：Reducing false positives is crucial in lung nodule detection since an excessive false positive can challenge radiologists' interpretation. In this research, we collected the medical exam (ME) dataset, whose diagnostic imaging is characterized by small nodules (<6mm) and blurriness (ground glass opacity, GGO). To account for overall sensitivity after false-positive reduction, we propose a 3D multi-scale channel-spatial attention model with multi-descriptor-based refinement, consisting of three sub-networks: the 3D multi-scale channel-spatial attention (MCSA), the 3D cross-attention, and the multi-descriptor-based classification (MDC). The 3D MCSA primarily comprises 3D multi-scale zoom-in and zoom-out streams with a channel-spatial attention module to enhance the quality of low-gradient nodules and facilitate feature extraction from diverse receptive fields, while the 3D cross-attention fuses feature maps from diverse receptive fields and generates the feature descriptor. The MDC utilizes multi-feature descriptors for the final 3D candidate nodule decision. Our model achieves sensitivities of 84.0 % and 97.3 % on the ME and LUNA16 candidate datasets under 2 FPs/scan, respectively.
查看更多>>摘要：The branch-and-bound (B&B) procedure is one of the most frequently used methods for solving the resource-constrained project scheduling problem (RCPSP) to obtain optimal solutions and has a rich history in the academic literature. Over the past decades, various variants of this procedure have been proposed, each using slightly different configurations to search for the optimal solution. While most of the configurations perform relatively well for many problem instances, there is, however, no known universal best B&B configuration that works well for all problem instances. In this work, we propose two problem transformation-based machine learning classification methods (binary relevance and classifier chains) to automatically detect the best-performing branch-and-bound configuration for the resource-constrained project scheduling problem. The proposed novel learning models aim to find the relationship between the project characteristics and the performance of a specific B&B configuration. With this obtained knowledge, the best-performing B&B configurations can be predicted, resulting in a better solution. A comprehensive computational experiment is conducted to demonstrate the effectiveness of the proposed classification models and the performance improvements over three categories of methods from the literature, including the latest branch-and-bound configurations, the state-of-the-art classification models in project scheduling, and commonly used clustering algorithms in machine learning. The results show that the proposed classification models can enhance solution quality for the RCPSP without changing the core components of existing algorithms. More specifically, the classifier chains method, when combined with the Back-Propagation Neural Network algorithm, achieves the best performance, outperforming binary relevance, which demonstrates the impact of label correlation on the performance. The experiments also demonstrate the merits of the proposed model in improving the robustness of the solutions. Furthermore, these findings not only highlight the potential of the classification models in detecting best-performing B&B configurations, but also emphasize the need for future work and development to further improve the performance and applicability of these models.
查看更多>>摘要：Classical queueing models fail to properly staff non-conventional call centers with complex internal structures. This is either due to the difficulty of finding suitable models whose underlying assumptions hold, or due to certain elements of the call center not being modeled such as caller patience times. Relay call centers, service providers that connect two different interested parties with one another through telecommunication channels, present a prime example of non-conventional call centers. Working on the study case of a relay call center for the deaf community, Erlang C, one of the most commonly used call center staffing formulae, fails to generate agent staffing that meets our target performance criteria for quality of service. We propose a machine learning-based approach leveraging an available log of historical data. Upon comparing the proposed approach's capability of performance evaluation and agent staffing to that of the Erlang C model and a baseline data-driven model, results indicate our approach's staffing superiority. Considering uncertainty within the system variable predictions carried out prior to the staffing phase, our approach generates agent staffing which enables us to meet our global quality of service objective.
查看更多>>摘要：River flow forecasting is crucial for water resource management, flood prevention, and environmental sustain-ability. River flow forecasting is crucial for water resource management, flood prevention, and environmental sustainability. Despite the application of many deep learning models in river flow prediction, they often face challenges such as limited prediction durations and insufficient accuracy. In this study, we propose an integrated deep learning model based on multi-scale feature extraction to enhance the accuracy of long-term river flow forecasts. The model integrates a multi-scale feature extraction module and a context-aware module. The former is responsible for capturing diverse features of river flow at multiple scales, while the latter further analyzes and models these features. Together, these modules enhance the model's performance in long-term river flow forecasting. Experimental results on a river flow dataset, predicting the flow for the next 120 h, demonstrate that the proposed model maintains high accuracy, thus validating its effectiveness for long-term river flow prediction.
查看更多>>摘要：The public-private partnership (PPP) is a financing model for infrastructure and public services that involves both government and social capitalists. PPP projects have been introduced to improve rural residential environments, attracting research from numerous scholars. However, few studies have considered rural enterprises as a source of rural environmental pollution or constructed game models to examine related multi-agent collaborative governance mechanisms under the influence of various factors. To address these research gaps, this study establishes a tripartite game model involving local governments, social capitalists, and rural enterprises in rural residential environmental improvement PPP projects. The findings of this study show that the policy benefits and incentive expenditures of local governments, as well as the environmental protection costs and long-term benefits of rural enterprises, significantly influence the strategic choices of all participants. Specifically, when the policy benefits of local governments exceed economic rewards and the environmental protection costs for rural enterprises are relatively low, the system evolves to a state where all participants actively engage in these projects. Obviously, strengthening policy support from the central government and reducing pollution control costs for rural enterprises can effectively enhance cooperation efficiency. This study not only provides novel insights into rural residential environmental improvement but also offers implications for optimizing policy implementation and guiding social capitalists' and rural enterprises' behaviors.
查看更多>>摘要：Facial expression recognition (FER) plays a crucial role in numerous human-computer interaction systems. For the sake of precise recognition, existing methods often enhance the representational capacity of networks by designing complex network structures or incorporating additional facial information. However, due to redundancy among facial expression features, refining expression-related information to obtain highly discriminative expression features remains challenging. We propose a multi-task self-distillation method with coarse and fine grained labels for FER. To mine the sufficient expression-related information, we construct coarse-grained auxiliary branches that enhance the learning ability of the network based on the prior in the facial expression labels. To map coarse-grained features into a fine-grained feature space, feature alignment modules are then introduced. Then, refined self-distillation is constructed to transfer coarse-grained knowledge to fine-grained features, providing additional guidance for the extraction of discriminative features. Our proposed method achieves state-of-the-art performance on multiple FER benchmarks, demonstrating its superiority.
Jose CalderonDaniel AyalaRafael AyalaLuis Valencia-Cabrera...
127453.1-127453.13页
查看更多>>摘要：Satellite orbit propagation involves predicting a satellite's future position and velocity based on initial conditions. Traditional physical models, such as SGDP4, simplify the forces that act on the satellite to achieve high computational efficiency at the cost of reduced prediction accuracy, especially over longer time intervals where error accumulates. More sophisticated models like HPOP offer improved accuracy at the cost of high prediction times, rendering them unusable for realtime long-term predictions. Recent advancements have introduced machine learning techniques to refine these predictions and reduce errors. However, they often lack an analysis of model design choices, such as input feature selection and architectural configurations. Existing models do not incorporate features related to the state of celestial bodies, such as the positions of the Moon or Sun, which can influence the satellite's trajectory. This paper proposes a novel model that integrates such features at both the initial time and throughout the prediction interval, leveraging their potential impact on the orbit of the satellite. The model is based on a neural network architecture employing GRU layers for encoding sequential data about the celestial conditions. Our results demonstrate that the inclusion of these sequential features significantly reduces prediction errors. Additionally, we have evaluated a variety of design choices such as independent sub-models for specific spatial coordinates and time intervals, further enhancing performance. These innovations lead to substantial improvements in both short- and long-term orbit predictions, providing a more robust and accurate alternative for satellite orbit propagation.
NETL
NSTL
Elsevier