Applied Intelligence最新文献

Multigranulation rough set is composed of a set of granularities, providing a theoretical framework for solving problems from a multigranulation perspective. Feature selection aims to find the minimal set of attributes that does not compromise the overall classification capability. It has significant applications in the field of information processing. However, in practical application environments, the granularities in information systems often evolve dynamically over time. To address this scenario, an incremental feature selection algorithm for data with changing granularities in local multigranulation neighborhood covering rough sets is proposed. Firstly, the method of local related family is introduced, relationships between matrix operations of local related sets and those of approximate sets are discussed, and feature selection is studied using matrix methods. Subsequently, two matrix-based incremental feature selection algorithms are proposed for the cases where granularity structures in the data are added or deleted due to feature changes. Experiments on six datasets from UCI are then conducted to evaluate the performance of the proposed algorithms. The experimental results demonstrate that the two proposed incremental feature selection algorithms are highly effective.

Exemplar-free class-incremental learning recognizes both old and new classes without saving old class exemplars because of storage limitations and privacy constraints. To address the forgetting of knowledge caused by the absence of old training data, we present a novel method that consists of two modules, multi-view prototype balance and temporary proxy constraints, which are based on feature retention and representation optimization. Specifically, multi-view prototype balance first extends the prototypes to maintain the general state of the class and then balances these prototypes combining knowledge distillation and prototype compensation to ensure the stability and plasticity of the model. To alleviate the feature overlap, the proposed temporary proxy constraint sets the temporary proxies to lightly compress the feature distribution during each mini-batch of training. Extensive experiments on five datasets with different settings demonstrate the superiority of our method against the state-of-the-art exemplar-free class-incremental learning methods.

In order to reflect the nonlinear dynamics of quantum game map more precisely and improve the performance of current cryptosystem, in this paper, we first obtain the fractional four-dimensional quantum game chaotic map according to the fractional calculus. We then proceed to obtain the trajectory, the maximum Lyapunov index, the bifurcation diagram of the aforementioned map, and compare its dynamical behaviours with those of the four-dimensional quantum game chaotic map. Subsequently, we design a synchronisation control system for the proposed fractional system and apply the two systems to the field of information security. Finally, we undertake a comprehensive analysis of the encryption system, examining it from five distinct perspectives. As a result, the key space of proposed algorithm equals to (6.7 times 10^{165}) and the encrypted image’s information entropy is 7.9997. The NPCR and UACI for the difference attack analysis is (99.61%) and (33.44%) respectively accompanied with that the correlation coefficients is near to 0. The result indicates that the proposed algorithm can defense both known plaintext and chosen plaintext attacks which means that it is superior to other algorithms in most of the above aspects.

The Gaussian Estimation of Distribution Algorithm (GEDA) is a fundamental evolutionary algorithm widely applied to continuous optimization problems but often encounters premature convergence. While external archives have been introduced to mitigate this issue, they frequently misuse historical information, leading to suboptimal results. To address this, we propose an Adaptive Archive Exploitation for GEDA (AAE-GEDA). AAE-GEDA incorporates two key mechanisms: adaptive selection of archive quantities (ASAQ) and angle skewness-landscape (ASL) eigenvalue adaptation. ASAQ selectively utilizes a subset of solutions from the archive to improve the accuracy of covariance estimation, preventing the algorithm from being misled by outdated or irrelevant information. ASL dynamically adjusts the search range, ensuring a balanced trade-off between exploration and exploitation. Experimental results on the IEEE CEC2014 and CEC2017 test suites demonstrate that AAE-GEDA consistently outperforms state-of-the-art evolutionary algorithms.

Vision Transformer (ViT) has demonstrated excellent accuracy in image recognition and has been actively studied in various fields. However, ViT requires a large matrix multiplication called Attention, which is computationally expensive. Since the computational cost of Self-Attention used in ViT increases quadratically with the number of tokens, research to reduce the computational cost by pruning the number of tokens has been active in recent years. To prune tokens, it is necessary to set the pruning rate, and in many studies, the pruning rate is set manually. However, it is difficult to manually determine the optimal pruning rate because the appropriate pruning rate varies from task to task. In this study, we propose a method to solve this problem. The proposed pruning rate adjustment adjusts the pruning rate so that the training loss is converged by Gradient-Aware Scaling (GAS). In addition, we propose Variable Proportional Attention (VPA) for Top-K, a general-purpose token pruning method, to mitigate the performance loss due to pruning. For the CIFAR-10 dataset, several competitive pruning methods improve recognition accuracy over manually setting the pruning rate; eTPS+Adjust on Hybrid ViT-S achieves 99.01% Accuracy with -31.68% FLOPs. Furthermore, Top-K+VPA outperforms token merging when the pruning rate is large for trained ViT-L inference on ImageNet-1k and has superior scalability in the Accuracy-Latency relation. In particular, when Top-K+VPA is applied to ViT-L on a GPU environment with a pruning rate of 6%, it achieves 80.62% Accuracy on the ImageNet-1k dataset with -50.44% FLOPs and -46.8% Latency.

The existing knowledge graph completion model represents entity and description information by uniform fusion. The convolutional kernel has fewer sliding steps on a triplet matrix composed of entities and relations and does not obtain different-scale characteristics for entities and relations. In this study, a knowledge graph completion model based on weighted fusion description information and the transformation of the dimension and scale, EDMSConvKE, is proposed. First, the entity description information is obtained using the SimCSE model of comparative learning and then combined with the entity according to a certain weight to obtain an entity vector with a stronger expression ability. Second, the head entity, relation, and tail entity vectors are combined into a three-column matrix, and a new matrix is generated using a dimensional transformation strategy to increase the number of sliding steps of the convolution kernel and enhance the information interaction ability of the entity and relation in more dimensions. Third, the multi-scale semantic features of the triples were extracted using convolution kernels of different sizes. Finally, the model in this study was evaluated using a link-prediction experiment, and the model was significantly improved in the Hits@10 and mean rank (MR) indices.

Predictive modeling plays a crucial role in machine learning, data analysis, and statistics. In sports, predictive modeling methods have emerged to provide insights and evaluate performances based on key performance metrics. However, most existing models tend to focus on predicting only partial aspects of an event, such as the outcome, action type, or location, while neglecting the temporal factors involved. To address this gap, this study introduces the Transformer-Based Neural Marked Spatio-Temporal Point Process (NMSTPP) model, specifically designed for football event data. The NMSTPP model predicts a comprehensive set of future event components, including inter-event time, zone, and action. Additionally, it features a dependent prediction layers architecture to enhance model performance. The Holistic Possession Utilization Score (HPUS) metric is also proposed to evaluate the effectiveness and efficiency of possession periods in football based on the NMSTPP model. With open-source football event data, the NMSTPP model successfully predicted the aforementioned three components of future events, with an improvement of up to 4% overall and 9% for individual components compared to baseline models. The HPUS demonstrated a 0.9 correlation with existing performance metrics, highlighting its utility in performance evaluation. The NMSTPP and HPUS were applied to the Premier League to demonstrate their practical feasibility.

Collaborative recommendation systems have been widely used in various fields, such as movies, music and e-commerce. However, due to the natural openness of its ratings, it is vulnerable to shilling attacks. Shilling attacks greatly affect the accuracy and trustworthiness of recommendation systems, so we urgently need effective methods to counter shilling attacks. Some detection methods have been proposed previously. However, they mostly use manual feature extraction-based methods. These methods require specialized statistical knowledge to summarize user-specific rating patterns in user rating databases, which is very difficult. Thus, we propose a method called User Similarity-based Graph convolutional neural network for Shilling Attack Detection (USGSAD). This method achieves the purpose of detecting shilling attacks without using manual features. First, our method calculates user similarity by jointing both correlation and deviation of user rating behaviors. Second, we build a user relationship graph based on user similarity matrix and use graph embedding method to obtain user low-dimensional embedding vectors. Finally, we design a User Similarity Graph Convolutional Network (USGCN) to assign weights to aggregate user embeddings and predict the attackers in the recommender system. Adequate experiments on Amazon and MovieLens datasets show that our proposed method outperforms the baseline methods in detection performance.

A predefined-time sliding mode adaptive control method (PDTSMAC)for nonlinear system is proposed in the presence of parameters unknown, external disturbances and arbitrary initial values. Firstly, the expected trajectory of the system is extended to the arrival process with characters of predefined-time convergence and the accurate tracking process of completely tracking the desired trajectory, the design principle of extended trajectory is given; Then, an extreme learning machine (ELM) with exponential convergence of external weights is designed to compensate the uncertainties of the system, and a sliding mode adaptive controller with predefined-time convergence is constructed based on a predefined-time convergent sliding mode surface. The stability of the closed-loop system is proved theoretically. The simulation results show that the control strategy can ensure that the construction robot in arbitrary initial state converges to the extended desired trajectory within the predefined-time, and realizes the complete and accurate tracking of the preset desired trajectory, and the trajectory tracking error is less than 0.008.

With the global population on the rise, the frequency and severity of emergency events like fires and traffic accidents are becoming more frequent and severe. Attending to these emergencies demands valuable and limited resources, such as professionals and vehicles, so it is important to efficiently allocate them to regions that are more likely to require their services. However, the fact that emergencies can be related to spatial and temporal contexts makes resource allocation a highly complex task requiring specialized tools and techniques to exploit these relationships efficiently. This paper proposes an emergency event prediction solution using spatial segmentation and Long Short-Term Memory (LSTM) neural networks to model associations in space and time domains. We used data from real emergency occurrences in Florianópolis, Brazil, collected over five and a half years. Clustering algorithms combined with the silhouette metric were used to segment the time series in four different city regions. A comparison with traditional forecasting techniques and machine learning models showed that the LSTM network is consistent in its predictions and outperforms other approaches. Compared with a state-of-the-art reference employing LSTM, our solution leads to a 17.8% reduction in mean absolute error. Two methodologies for multi-step lookahead prediction are also presented and compared, showing that reusing the output of LSTM to predict future time steps is better than a full model retraining. To assess the generalizability of the model and proposed methodology, we applied the entire pipeline to new data from a different city. Our results demonstrate that models tailored to specific cities significantly outperform those trained on generalized datasets, highlighting the importance of localized training data.