Applied Intelligence最新文献

In real decision-making scenarios involving multi-agent games, such as intelligent unmanned systems, military confrontations, and autonomous navigation, the coupling of participant strategies and the incompleteness of perceived information make the accurate inference of dynamic game trajectories a critical and challenging task. To address this problem, this paper proposes a game trajectory modeling method that integrates a cross-attention mechanism with a moving diffusion process, termed CARM-Diff (Cross-Attention Autoregressive Moving Diffusion Model). The model combines autoregressive structures with cross-attention to capture the temporal evolution of sequences while explicitly modeling strategic interactions between agents. We design a lightweight feature extraction module and, leveraging the Markov property of diffusion models, introduce a deterministic evolution process of historical states to simulate noise, thereby enhancing the model’s capability to learn local temporal patterns. Meanwhile, a cross-attention mechanism is introduced in the reverse diffusion stage to guide the model in focusing on the opponent’s historical sequential behavior, enabling more precise capture of inter-agent influences. Furthermore, we design a residual gated trajectory modeling structure that fuses the agent’s own behavioral evolution with interaction effects induced by opponents. Gating factors are dynamically generated through multilayer perceptrons to achieve adaptive information fusion. We construct a dynamic trajectory dataset based on an underwater pursuit-evasion game to validate our approach, and the proposed CARM Diff framework is generalizable to a wide range of multi-agent interactive systems. Experimental results show that CARM-Diff outperforms mainstream baseline methods in both prediction accuracy and dynamic interaction modeling, demonstrating the effectiveness and practical potential of the proposed model.

Fuzzy min-max neural network (FMNN) realized based on hyperbox has been widely used in the field of pattern classification. However, the parameter dependency coming from hyperboxes still remains open. Specifically, the construction process of the hyperbox is influenced by the expansion coefficient, different coefficients result in varying final configurations of the hyperbox. To address this issue, this paper proposes a fuzzy reinforced hyperbox neural network(FRHNN). FRHNN employs an innovative multilayer structure to generate hyperboxes, and its construction process is no longer constrained by traditional expansion coefficients. Compared to traditional hyperbox layers, the model is divided into two phases: hyperbox initialization and hyperbox enhancement. In the initialization phase, a clustering algorithm is used to form the initial mixed hyperboxes. Subsequently, during the hyperbox enhancement phase, the multilayer structure will progressively segment the mixed hyperboxes until each hyperbox is transformed into a pure hyperbox. The hyperbox enhancement process of this model encompasses three key steps: judgement hyperbox types, determining hyperbox categories, and executing hyperbox segmentation. A comparative study also indicates that the proposed FRHNN leads to higher classification accuracy in comparison with some state-of-the-art fuzzy min-max neural networks in tackling data classification.

Accurate emotion recognition in dialogue depends on making effective use of conversational context, yet many multimodal systems under-utilize this contextual information and misclassify nuanced speech. We introduce Multimodal Contextual Transformer Augmented Fusion (MCTAF), a lightweight context-sensitive transformer-based multimodal model that encodes speech and transcripts with Bi-GRUs and treats the last K utterances as a separate modality for contextual information. The context module summarizes the most recent K preceding utterances into a single context vector. This vector is fed to the fusion transformer as its own information stream, so cross-attention can relate the current text and audio to dialogue history without simply concatenating features. All modality vectors operate within a shared 128-dimensional space and are integrated through twelve directed cross-modal attention over text, audio, context modality, and an early-fusion feature. Trained end-to-end with Adam, MCTAF attains 89.9% accuracy on IEMOCAP and 88.3% on MELD, boosting weighted F1 by up to 3% and accuracy by up to 4% points over strong state-of-the-art methods. Ablation studies show that removing the context branch lowers weighted F1 by 3 to 4 points, demonstrating the importance of modeling explicit contextual module for improved emotion recognition. These results on benchmark datasets show that the inclusion of a dedicated contextual module yields consistent gains in both dyadic and multi-party conversations with modest computational cost.

The high-dimensional decision space of large-scale multi-objective optimization poses challenges to evolutionary algorithms, which leads to potential trapping in local optima. This paper proposes a brain information decomposition mechanism inspired large-scale multi-objective evolutionary algorithm (IDLMEA). Three strategies are developed. 1) The redundant information subpopulation division strategy selects high-quality solutions using shift-based density estimation, which determines the global optimal information. 2) The synergistic information subpopulation division strategy identifies exploration direction information by dividing solutions using local sensitive hashing. 3) The redundant and synergistic information-based reproduction strategy generates offspring for effective exploration of the search space. The Friedman test values of IDLMEA for inverted generational distance on LSMOP and IMF benchmarks outperform ten competitors by at least 25.64% and 4.04%, respectively.

The cyber security situation rating (CSSR) method and its stability analysis is key problems in nowadays cyber security. However, the existing CSSR methods do not take into account to the issues such as the fuzziness and hesitation of the grades of attributes, the non-compensation of information among attributes, the internal differences of attributes and the stability of CSSR method. A new piecewise variable weight rating (PVWR) method for CSS is proposed, and then the stability analysis for the PVWR method is researched. Based on the description of the CSS rating problem, linear membership functions (LMFs) and linear non-membership functions (LNMFs) of attributes with four categories grade threshold are constructed. The definition of PVWR function is proposed to more accurately reflect the characteristics of CSS, which is constructed by continuous piecewise function. A novel stability analysis of the PVWR method is proposed, and the attributes’ weight ranges under the condition of the current grade assessment is given when the CSSRM results is stable. The proposed method not only offers a novel method to identify the key impact attributes, but also provides theory basic for judging whether CSSR results have changed.

In the real world, there is a significant amount of incomplete data, making attribute reduction for incomplete data a critical issue. Many existing reduction methods mainly study the relationship between conditional attributes and decision attributes, and often lack the correlation between conditional attributes. In this paper, first, incomplete Euclidean distance and global similarity are proposed, and correspondingly, fused fuzzy similarity relations are constructed to characterize the fuzzy similarity relations between objects with missing values, and to generate an object relation matrix. Second, incomplete cosine similarity is proposed to characterize the similarity between conditional attributes, and an attribute relation matrix is generated to describe the correlation between conditional attributes. Finally, we construct the bidirectional fuzzy similarity discriminability (BFD) and the corresponding reduction algorithm on this basis. Under the KNN classifier, the average lead compared to other state-of-the-art algorithms is 6.44%, and this value reaches 6.67% under the CART classifier and demonstrate the effectiveness of the algorithm through experiments.

In response to the issues of incomplete local data degradation characterization and multi-scale, non-stationary characteristics exhibited during the degradation process of rotary machinery vibration signals, a remaining useful life (RUL) prediction method based on trend encoding and multi-scale spatio-temporal feature fusion is proposed. Firstly, a trend encoding method is introduced to compensate for the missing temporal information and long-term degradation information in vibration signal samples, enhancing the degradation characterization ability of local data. Subsequently, a soft threshold self-attention mechanism is proposed for the adaptive fusion of trend-encoded features and vibration signal features, to prevent imbalanced weight distribution. Finally, a spatio-temporal feature fusion network, MACNN-Informer, is designed. It possesses multi-scale spatial feature extraction capabilities and can effectively capture long-distance dependencies in sequential features, thereby better revealing the degradation characteristics of vibration signals at different degradation stages. Experimental results show that, compared with methods such as MSCNN, the proposed method–despite being slightly slower in inference speed–achieves the lowest prediction error and oscillation amplitude, making it well-suited for RUL prediction of critical rotating machinery components such as rolling bearings.

Multi-view clustering has important applications in machine learning and computer vision, but existing methods face three major limitations: insufficient consideration of the differences among views, failure to capture the complex relationships among views, and neglect of the interdependence of embeddings and clustering labels. To overcome these limitations, this paper proposes a novel dependency-driven spectral embedding based multi-view clustering (DDSE), which introduces a dual-layer learning system. On the global layer, DDSE highlights the differences among views by selectively retaining the discriminative features of each view and adjusting their weights; on the local layer, Grassmann manifolds are used to maintain the topological information among views and improve clustering adaptability. By learning a unified embedding from the Reproducing Kernel Hilbert Spaces, DDSE can capture high-order nonlinear dependencies among views and avoid information loss by generating a discrete indicator matrix. In addition, this paper derives an efficient optimization scheme to improve the performance of the proposed method. Multiple rounds of experiments on ten datasets verify the advantages of this method over other state-of-the-art methods.

Anomaly detection in industries is critical to ensure safety, reliability, and product quality. Industrial processes often produce high-dimensional multivariate time series (MVTS) data from sensors and actuators, making the detection of subtle or complex anomalies increasingly challenging. In this work, we introduce the novel application of curriculum learning (CL) to deep MVTS anomaly detection, a strategy inspired by the human learning process that begins training with easier examples before progressing to more difficult ones. We propose two CL frameworks: a data-based curriculum, which ranks training samples by signal complexity (e.g. noise level, window size), and a model-based curriculum, in which the knowledge learned from a simple long short-term memory encoder-decoder model is used to initialize a more advanced multi-scale convolutional recurrent encoder-decoder model. Furthermore, we present the first use of system signature matrices as an MVTS representation that captures spatial-temporal dependencies in textile process data for collective anomaly detection. We perform comprehensive evaluations on the SWaT benchmark dataset and a dataset collected from real-world textile processes. Our empirical results demonstrate the possibility of designing curriculum-trained models that outperform standard training baselines, achieving higher F1 scores while offering more structured training dynamics. Notably, the data-based CL strategy consistently yields better performance than non-curriculum baselines. This study represents the first systematic adaptation of CL to industrial anomaly detection and establishes a foundation for more structured training paradigms in MVTS anomaly detection.

In the context of online time series forecasting, experience replay (ER) is a crucial technology to mitigate catastrophic forgetting arising from concept drift. ER uses historical data sampled from a buffer to update model parameters along with current data incrementally. However, existing ER methods overlook buffer management, which is certainly important due to the space constraints and dynamic data distributions. To fill this gap, we propose the three-way experience replay (TWER). In contrast to existing ER methods, TWER is a comprehensive buffer management mechanism that involves admission, sampling, and eviction to improve the accuracy of online prediction. To capture samples indicative of concept drift, we design the admission mechanism based on three-way decision (TWD) rules deduced from historical and future data. By integrating Bayesian decision theory and rough set theory, TWD can provide the option of deferment action, enabling a more informed judgment on concept drift. To preserve the historical data with rare patterns, we enhance the sampling and eviction mechanisms by introducing three-way clustering (TWC), which can identify fringe samples typically missed by existing methods. Leveraging these mechanisms, TWER can adaptively manage data within the buffer, thus enhancing the prediction under concept drift. Empirical results on five real-world datasets show that TWER reduces online prediction error by more than 12% in terms of MSE compared with the existing ER methods such as DER++, GDumb, and memory select.