Information Sciences最新文献

Vision Transformer (ViT) has emerged as a pivotal model for a variety of visual tasks, surpassing convolutional neural networks by a substantial margin. However, the performance of ViT is seriously impaired by intensive computational and storage costs requirements, posing significant barriers for real-world applications or deployment on resource-constrained edge devices. To address this limitation, compressing the ViT to accelerate its inference at no appreciable degradation of vision performance has attracted widespread attention. Although there are some studies on accelerating ViT, they seldom consider resource constraints and multi-criteria decision making in the process. This article formulates ViT pruning as a large-scale constrained multi-objective optimization problem, and proposes a patch pruning framework for accelerating ViT, called EvolutionViT, based on the developed multi-objective optimization model. EvolutionViT can effectively tradeoff between computational cost and performance under resource constraints, automatically searching for solutions while optimizing two conflicting objectives. In particular, exploiting the knee solution and boundary solutions to directly guide the entire evolutionary process, EvolutionViT can efficiently identify a knee solution that satisfies the resource constraints, which in turn avoids the manual search for a good trade-off. To verify and evaluate our proposed method, we compare EvolutionViT with a number of representative ViT models on the ImageNet dataset. The comprehensive simulation results show that the proposed EvolutionViT demonstrates a competitive advantage compared to peers, with significantly reduced computational expense at the cost of slightly degraded performance.

Multi-objective evolutionary algorithms (MOEAs) are widely employed to tackle multi-objective optimization problems (MOPs). However, the choice of different crossover operators significantly impacts the algorithm's ability to balance population diversity and convergence effectively. To enhance algorithm performance, this paper introduces a novel multi-state reinforcement learning-based multi-objective evolutionary algorithm, MRL-MOEA, which utilizes reinforcement learning (RL) to select crossover operators. In MRL-MOEA, a state model is established according to the distribution of individuals in the objective space, and different crossover operators are designed for the transition between different states. Additionally, in the process of evolution, the population still exhibits inadequate convergence in certain regions, leading to sparse areas within the regular Pareto Front (PF). To address this issue, a strategy for adjusting weight vectors has been devised to achieve uniform distribution of the PF. The experimental results of MRL-MOEA on several benchmark suites with a varying number of objectives ranging from 3 to 10, including WFG and DTLZ, demonstrate MRL-MOEA's competitiveness compared to other algorithms.

Conventional machine learning algorithms face significant limitations when dealing with high-dimensional data. Besides, deep learning models often require substantial computational resources and have a high processing time despite their excellent performance. Hence, this paper proposes an expanded stochastic configuration network and a self-organizing hierarchical incremental learning (SHIL) framework to overcome these challenges. Specifically, this study introduces a novel supervised hierarchical clustering tree based on the minimum redundancy maximum correlation algorithm, which mines internal data structures to construct diverse hierarchies. Subsequently, by exploiting the parent-child node relationships in the tree structure, SHIL defines the maximum number of nodes as the switching condition between levels uses the supervisory mechanism as the parameter selection criterion, and adopts the tolerance error as the termination criterion for the training. Furthermore, the universal approximation property of the SHIL framework is provided. The proposed SHIL framework is validated on several benchmark datasets, image datasets, and industrial robot cases, with the corresponding experimental results demonstrating that SHIL significantly improves computational efficiency and ensures high accuracy.

The multimodal emotion recognition in conversation (ERC) task presents significant challenges due to the complexity of relationships and the difficulty in achieving semantic fusion across various modalities. Graph learning, recognized for its capability to capture intricate data relations, has been suggested as a solution for ERC. However, existing graph-based ERC models often fail to address the fundamental limitations of graph learning, such as assuming pairwise interactions and neglecting high-frequency signals in semantically-poor modalities, which leads to an over-reliance on text. While these issues might be negligible in other applications, they are crucial for the success of ERC. In this paper, we propose a novel framework for ERC, namely multimodal graph learning with framelet-based stochastic configuration networks (i.e., Frame-SCN). Specifically, framelet-based stochastic configuration networks, which employ 2D directional Haar framelets to extract both low- and high-pass components, are introduced to learn the unified semantic embeddings from multimodal data, mitigating prediction biases caused by an excessive reliance on text without introducing an unnecessarily large number of parameters. Also, we develop a modality-aware information extraction module that is able to extract both general and sensitive information in a multimodal semantic space, alleviating potential noise issues. Extensive experiment results demonstrate that our proposed Frame-SCN outperforms many state-of-the-art approaches on two widely used multimodal ERC datasets.

This study proposes a stacking fuzzy classifier with stochastic configuration-based learning that can achieve higher training and testing performances and sound interpretability of fuzzy rules. By using the understandable first-order Takagi–Sugeno–Kang fuzzy system, we initially stack each successive subclassifier on both the remaining misclassified training data and the corresponding outputs of the previous subclassifier. Subsequently, a Stacking Fuzzy Classifier with Fully Interpretable and Short fuzzy Rules (FISR-SFC) further improves its prediction by linearly aggregating the outputs of all the subclassifiers. FISR-SFC trains each subclassifier using the proposed stochastic configuration-based learning procedure to utilize its training excellence on gradually smaller misclassified training data and simultaneously maintain the full interpretability of each subclassifier. Experimental results on twelve benchmarking datasets reveal that FISR-SFC is at least comparable to and even better than the comparative classifiers in terms of average testing accuracy/G-mean and/or short rules with full interpretability.

In recent years, there has been rapid development in video anomaly detection (VAD). The previous methods ignored the differences between normal videos and only emphasized learning the commonalities of normal videos. In order to improve the performance of anomaly detection, we delve into the spatial distribution of normal video features and utilize their differences for clustering, leading to more minor reconstruction errors for normal videos and more significant reconstruction errors for abnormal videos. To achieve this goal, we introduce a Multi-Cluster Memory Prototype framework (MCMP) for VAD, which explores the coarse-grained and fine-grained information from video snippet features simultaneously to learn a memory prototype, thereby significantly improving the ability to discriminate abnormal events in complex scenes. First, a video feature clustering method that employs contrastive learning is introduced to group samples sharing similar fine-grained features. Second, the memory mechanism is used to capture the feature distribution of normal samples. Lastly, the Gaussian filter feature transformation method is introduced to make normal and abnormal features more distinguishable. The frame level AUC of MCMP on ShanghaiTech and UCF-Crime benchmark datasets has increased by 1.26% and 0.45% compared to state-of-the-art methods. Our code is publicly available at https://github.com/WuIkun5658/MCMP.

This paper investigates the finite-time quasi-projective synchronization (FTQPS) issue of fractional-order reaction-diffusion neural networks (FORDNNs). To the best of our knowledge, this paper introduces the concept of FTQPS for the first time. First, an integral-type Lyapunov function is constructed relying on the characterization of the reaction-diffusion term and some inequality methods. Subsequently, the nonlinear feedback control strategy is designed to achieve the FTQPS goal and some sufficient conditions are obtained to guarantee FTQPS of FORDNNs. Further, the system's synchronization speed is measured by estimating the settling time. It should be noted that the above control strategy is also applicable to conventional integer-order reaction-diffusion neural networks with time delays. Finally, a numerical example is used to illustrate the validity of the theoretical analysis presented.

The graph model for conflict resolution (GMCR) is a useful tool for modeling and analyzing conflicts. In a conflict, the decision maker (DM)’s evaluation of feasible states is often influenced by multiple attributes. When in different feasible states, DMs may assign different importance to each attribute. Therefore, this paper applies multi-attribute evaluation (MAE) to GMCR and proposes the MAE-based stability definition. In addition, due to differences in relationships and opinions among DMs, opponents will behave differently in response to the action of the focal DM. To analyze the heterogeneous behavior of opponents, this paper proposes a heterogeneous behavior analysis method based on social network analysis (SNA) and opinion similarity. Then, the MAE-based mixed stability definitions are proposed to perform the stability analysis considering heterogeneous behaviors. Finally, this paper applies the proposed method to the Elmira contamination conflict and makes a sensitivity analysis to prove the validity of the proposed method.

Geothermal energy stands out as an exceptional renewable resource for power generation, offering a consistent power production without the intermittency issues. Despite its potential to deliver a consistent supply of electricity on demand, geothermal adoption is hindered due to substantial costs. Utilising the most effective drilling method can alleviate this challenge by boosting efficiency and reducing operational costs. The primary goal of this study is to identify the best drilling method for extracting heat from geothermal reservoirs. This optimised approach facilitates better access to geothermal reservoirs, leading to increased heat recovery rates and improved project viability. Traditional methods often fall short in evaluating optimal drilling alternatives due to uncertainties. To address this, our research introduces an innovative paradigm that integrates novel T-Spherical Hesitant Fuzzy Rough ($T-\mathrm{SHFR}$) set, method for the removal effects of criteria with a geometric mean and ranking alternatives with weights of criterion hybrid Multiple Criteria Decision-Making (MCDM) techniques. By leveraging the novel $T-\mathrm{SHFR}$ concept, our approach allows for a comprehensive assessment of various factors. This holistic evaluation ensures an exhaustive comprehension of the decision-making environment. The study reveals that reservoir characteristics play a significant role in selecting a sustainable drilling alternative. Furthermore, directional drilling appears as the most promising method with higher energy yields followed by slim hole drilling. The robustness and credibility of these findings are established through sensitivity and comparative analyses, indicating the potential applicability of this MCDM method to analogous challenges in different contexts. The findings of the ranking techniques were validated using Spearman's rank correlation coefficient, which revealed a positive and notable correlation. This research will empower stakeholders to make informed decisions, thereby enhancing the overall efficiency and sustainability of geothermal energy projects.

In the midst of today's intricate and ever-changing natural and social landscape, this study is committed to proposing a novel multi-source information fusion method for assessing the spatial resilience of urban emergency evacuation and rescue. Its overarching aim is to uncover latent challenges and contribute to the enhancement of a city's capacity to respond to emergencies. To achieve this, we have devised a comprehensive assessment indicator system, capable of not only quantifying a city's spatial resilience but also offering indispensable guidance for urban emergency evacuation and rescue spatial planning. Furthermore, we have introduced an innovative evaluation methodology framework. This framework includes the establishment of the basic probability assignment (BPA) model, a pioneering method for generating BPA for observed values, and the novel application of hierarchical weighting and fusion techniques. By extending the Dempster-Shafer theory, we have not only enhanced the method's ability to express and integrate uncertain information but also significantly improved the precision of the evaluation. Ultimately, we conducted a quantitative assessment using Shenzhen, China, as a case study, identifying existing issues and proposing highly-targeted improvement strategies. These research findings not only provide robust support for the augmentation of urban emergency capabilities in Shenzhen but also offer pioneering insights for the quantitative assessment and advancement of emergency spatial resilience in cities across the globe.