Applied Intelligence最新文献

This paper presents two hyper-heuristic approaches for solving a facility location problem called reliable p-median problem with at facility service (RpMF). In RpMF, service is provided to customers at the facility locations and it is closely related to the p-median problem. p-median problem is concerned with locating p-facilities while minimizing the total distance traveled by the customers to the corresponding nearest facilities and it is an (mathcal{N}mathcal{P})-hard problem. But according to the p-median problem, it doesn’t consider the possibility of facility failures. On the other hand, RpMF assumes that facilities can fail and the customers assigned to that facility do not know about the facility failure till they reach the facility for service. So, the customers have to travel from failed facilities to other functioning facilities to receive service. RpMF deals with locating p facilities to minimize the cost of serving the customers while considering facility failures. We have proposed two multi-start hyper-heuristic based approaches that are based on greedy and random selection mechanisms to solve the RpMF. The solutions obtained through hyper-heuristics are improved further via a local search. The two proposed hyper-heuristic approaches are evaluated on 405 RpMF benchmark instances from the literature. Experimental results prove the effectiveness of the proposed approaches in comparison to the state-of-the-art approaches available in literature for the RpMF.

Learning control policies for real-world robotic tasks often involve challenges such as multimodality, local discontinuities, and the need for computational efficiency. These challenges arise from the complexity of robotic environments, where multiple solutions may coexist. To address these issues, we propose Composite Gaussian Processes Flows (CGP-Flows), a novel semi-parametric model for robotic policy. CGP-Flows integrate Overlapping Mixtures of Gaussian Processes (OMGPs) with the Continuous Normalizing Flows (CNFs), enabling them to model complex policies addressing multimodality and local discontinuities. This hybrid approach retains the computational efficiency of OMGPs while incorporating the flexibility of CNFs. Experiments conducted in both simulated and real-world robotic tasks demonstrate that CGP-flows significantly improve performance in modeling control policies. In a simulation task, we confirmed that CGP-Flows had a higher success rate compared to the baseline method, and the success rate of GCP-Flow was significantly different from the success rate of other baselines in chi-square tests.

This study presents a novel approach to enhance the accuracy of automatic classification systems for airborne pollen particles by integrating domain adaptation techniques. Our method incorporates expert-verified measurements into the convolutional neural network (CNN) training process to address the discrepancy between laboratory test data and real-world environmental measurements. We systematically fine-tuned CNN models, initially developed on standard reference datasets, with these expert-verified measurements. A comprehensive exploration of hyperparameters was conducted to optimize the CNN models, ensuring their robustness and adaptability across various environmental conditions and pollen types. Empirical results indicate a significant improvement, evidenced by a 22.52% increase in correlation and a 38.05% reduction in standard deviation across 29 cases of different pollen classes over multiple study years. This research highlights the potential of domain adaptation techniques in environmental monitoring, particularly in contexts where the integrity and representativeness of reference datasets are difficult to verify.

Enhancing the integrity of the information security infrastructure requires the monitoring and analysis of anomalous network activities. And due to the network ecosystem's increased diversity and complexity as a result of information technology's rapid growth, classic intrusion detection techniques are no longer adequate for identifying and evaluating network anomaly patterns from a variety of integration and channel viewpoints. Meanwhile, the class imbalance problem associated with intrusion detection datasets limits classifiers' ability to recognize minority classes. To improve the detection rate of minority classes while ensuring efficiency, we propose a multi-channel intrusion detection model based on CNN_LSTM, referred to as ENS_CLSTM.The model that is being provided resamples the data using the sliding window approach and information entropy technology in order to balance the amount of normal and abnormal classes. The spatial features of the data are retrieved using a Convolution Neural Network (CNN), while the temporal features are extracted using a Bidirectional Long-Short Term Memory (Bi_LSTM), after integrates the dual-channel features stream into the final Deep Neural Network (DNN). The advantages of the proposed model are verified using the NSL-KDD,UNSW-NB15,CICIDS2017,CSE-CIC-IDS-2018 and ISCX-IDS2012 datasets. According to the experimental results, an accuracy of 99.67% was attained on the UNSW-NB15 dataset and 99.997% on the NSL-KDD dataset. Furthermore, on the CICIDS2017, CSE-CIC-IDS-2018, and ISCX-IDS2012 datasets, respectively, accuracy rates of 99.9997%, 99.998%, and 99.74% were attained.The ENS_CLSTM model can effectively improve the detection performance and generalization ability when compared to the findings of current studies.

Obstructive sleep apnea-hypopnea syndrome (OSAHS) is a prevalent chronic disorder that affects sleep quality and general health. The current diagnostic methods, primarily polysomnography (PSG), are laborious. Furthermore, audio-based methods for diagnosing OSAHS face limited sample sizes and neglect patients’ physiological signs and medical histories. To address these challenges, we introduce a data-driven framework called DFNet, which also considers patients’ medical histories and health indicators. DFNet incorporates an automated audio segmentation- and labeling-based preprocessing procedure to reduce expert annotation costs and subjective errors. We employed random convolutional kernels based on receptive fields for audio feature extraction purposes. These kernels captured both local and global features within the input audio. Additionally, for the first time, we introduced a medical language model that utilizes patients’ medical histories and physiological information as covariates to enhance features. We extensively validated DFNet on an OSAHS dataset obtained from a collaborative university hospital. Our framework classified patients into four categories according to their OSAHS severity: normal, mild, moderate, and severe. DFNet achieved state-of-the-art performance, with a four-class accuracy of 84.12%. DFNet offers a large-scale and cost-effective screening approach for diagnosing OSAHS, reducing the labor requirements of diagnosis. Our code is available at https://github.com/testlbin/DFNet.

Knowledge graphs (KGs) is an associated network composed of semantic relationships. The goal of the knowledge graph question answering (KGQA) is to provide answers to natural language questions based on KGs. Multi-hop KGQA requires reasoning on multiple edges of KGs to get the correct answer. However, KGs are usually incomplete, with numerous missing relationships in reality, which brings challenges to KGQA, especially for multi-hop KGQA. In this work, we propose an efficient approach for multi-hop KGQA. To capture more comprehensive features on incomplete KGs, we utilize Tucker Entity Relation (TuckER) decomposition for link prediction on the binary tensor representation of KGs and train a knowledge graph embedding (KGE) model and apply the learned representation for downstream QA tasks. We employ a pre-trained language model to assess the relevance scoring of questions and each node after subgraph retrieval. Additionally, we introduce a link scoring strategy based on the triple scoring function to address the limitations of solely relying on KGE for answer scoring. Through extensive experiments conducted on multiple benchmark datasets, we demonstrate the effectiveness of our proposed model in facilitating multi-hop QA reasoning on incomplete KGs.

The performance of machine learning algorithms can be optimized through the implementation of methodologies that facilitate the development of autonomous and adaptive behaviors. In this context, the incorporation of goal-oriented analysis is proposed as a means of effecting a transformation in the behavior of traditionally “passive" algorithms, such as Random Forest, through the endowment of proactivity. The aforementioned analysis, represented using the i* modeling language, identifies strategies that increase the diversity of generated trees and optimize their total number while preserving the original model’s effectiveness. In addition to the outcomes achieved, it is crucial to highlight that the goal-oriented methodology plays a pivotal role in the development and comprehension of novel algorithmic variants. Based on this analysis, two proactive variants were designed: the Proactive Forest and the Progressive Forest. These variants balance simplicity and effectiveness, maintaining the original algorithm’s performance while exploring more efficient configurations. This work introduces new variants of the Random Forest algorithm and demonstrates the potential of goal-oriented analysis as a methodology for guiding the design of more adaptive and effective algorithms.

With the rapid development of the Internet industry, a large amount of streaming data with significant application value will be generated on the Internet. The distribution of stream data is evolving over time compared to traditional data, posing a significant challenge in the learning process from streaming data. In order to adapt the change of data distribution, concept drift detection methods are proposed to pinpoint when the concept drift occurs. Most existing drift detection methods, however, overlook the improvement of the current classifier and the influence of noise data on drift detection. This oversight leads to a decrease in the effectiveness of drift detection. In this paper, we propose a novel adaptation drift detection method to overcome the shortcomings of previous algorithms, such as error detection and lack of anti-noise capability. Meanwhile, stream computing and parallel computing are used to enhance the efficiency of our algorithm. The results of a simulation experiment on 9 synthetic stream data and 6 real-world stream data, all exhibiting concept drift, demonstrate that our method is more effective in handling concept drift compared to other state-of-the-art methods.

Acne grading is a critical step in the treatment and customization of personalized therapeutic plans. Although the knowledge distillation architecture exhibits outstanding performance on acne grading task, the impact of non-label classes is not considered separately, resulting in low distillation efficiency for non-label classes. Such insufficiency will cause the misclassification of the acne images located on the edge of the decision boundary. To address this issue, a novel method named Adaptive Decoupled Knowledge Distillation (ADKD) which considers the uniqueness of the acne images is proposed. In order to explore the influence of non-label classes and enhance the model’s distillation efficiency on them, ADKD splits the traditional KD loss into two parts: non-label class knowledge distillation (NCKD), and label class knowledge distillation (LCKD). Additionally, it dynamically adjusts the NCKD based on the distance between the sample and each non-label class. This allows the model to allocate different learning intensities to various non-label classes, reducing the overrecognition of classes near the sample and the underrecognition of distant classes. The proposed method enables the model to better learn the fuzzy features between acne images, and more accurately classify the samples located on the decision boundary. To verify the proposed method, extensive experiments were carried out on ACNE04 dataset, ACNEHX dataset, and DermaMnist dataset. The experimental results demonstrate the effectiveness of this method, and its performance surpasses that of current state-of-the-art (SOTA) method.