Vision-and-language navigation (VLN) requires agents to follow natural language instructions and navigate in previously unseen environments. Large language models (LLMs) bring strong reasoning and generalization abilities to VLN. However, existing LLM-based methods still suffer from information loss in vision-to-text conversion, rigid prompting, or manually designed reasoning steps, as well as weak closed-loop control, leading to error accumulation. We propose DeepVLN, an LLM-based VLN framework that enables autonomous chain-of-thought (CoT) reasoning and adaptive navigation. DeepVLN first adapts an open-source LLM to VLN via a three-stage supervised fine-tuning pipeline, and then further optimizes its closed-loop policy with reinforcement learning, encouraging robust feedback use and online error correction without predefined reasoning templates. Additionally, an API-based collaborative reasoning module enables a lightweight local agent to selectively query a stronger cloud LLM under high uncertainty, thereby balancing performance and computational cost. Experiments on R2R, RxR, and REVERIE show that DeepVLN achieves competitive or superior results to strong VLN-specific and LLM-based baselines, with higher success rates and path efficiency in most settings. These results demonstrate the effectiveness of equipping LLMs with autonomous, closed-loop reasoning for embodied navigation.

Feature Selection (FS) is essential for high-dimensional data analysis. It results in compact and information-preserving representations that offer superior discriminability, improved computational efficiency, and greater interpretability. However, existing methods struggle to model complex inter-feature couplings. Reinforcement Learning (RL)-based FS approaches also face large action spaces, ambiguous credit assignment, and redundant agent behaviors. To address these issues, we propose Elastic Agents for Cooperative Feature Selection (EAC-FS), a scalable and interpretable Multi-Agent Reinforcement Learning (MARL) framework. EAC-FS treats FS as a dynamic cooperative process. Firstly, it introduces elastic agents that adaptively transfer feature ownership to reduce redundant computation and promote specialization. Furthermore, a local observation encoder captures structural information within each feature group, while a hierarchical credit assignment scheme separates global objectives from local contributions. A parallel transfer policy further coordinates feature routing across agents. Ultimately, EAC-FS is the first unified MARL formulation in which feature grouping, transfer, and selection evolve jointly within a single sequential decision process. Extensive experiments on nine benchmark datasets show that EAC-FS outperforms both classical and RL-based methods in accuracy, robustness, and cross-classifier generalization.

One-class classification (OCC) is a classification problem in which the training data contains only one class. The one-class support vector machine (OCSVM) is one of the most competitive OCC algorithms. However, OCSVM has scalability issues with large-scale datasets. This paper proposes the acceleration strategy of OCSVM. The idea is to decompose the dataset into samples and train OCSVM models for single data points. Subsequently, ensemble learning is applied to combine all models to compute the OCSVM model for the dataset. In addition, further acceleration is achieved through a data-reduction strategy with an OCSVM model trained on the average of the training samples. The experiment compared the proposal and traditional OCSVM using the Python package. The proposed strategy is faster than traditional OCSVM, while achieving similar classification results. Moreover, the proposed strategy can create one-to-one correspondence between samples and models. Source code is uploaded at https://github.com/ToshiHayashi/ODSVM

In the risk evaluation of intelligent ships during navigation, relying solely on the FMEA evaluations of a limited number of domain experts may be constrained by incomplete professional knowledge and susceptible to subjective bias. Large language models (LLMs) offer cross-domain synthesis but require mechanisms to control reliability and align with human judgment. This paper proposes a group Experts-LLMs collaborative decision making for FMEA Risk E-valuation, which involves the role-conditioned, the prompt-engineered LLM agents generating Failure Mode-Risk Factor assessments and the targeted mitigation suggestions. Then, a multi-round consensus model with personalized tolerance reconciles heterogeneous opinions to improve the group Experts-LLMs cooperation. Hence, the reliability analysis is proposed for group Experts-LLMs cooperation by the threefold: (1) The Robustness Analysis, (2) The Ranking Consistency Analysis, and (3) The Expert Scoring Deviation Analysis. The proposed model is validated through a case study, followed by comparative experiments and sensitivity analyses to further assess its effectiveness and robustness.

Machine Learning (ML) techniques have been applied to various problems. However, applying ML to ML models is an unexplored direction. For this purpose, this paper considers a meta-classification of one-class classification (OCC) models, because all ML models could be approximated as OCC models. The proposal represents OCC models as normality rankings and classifies them using nearest-neighbor and ranking-correlation metrics. The experiment classifies OCC models, where classes correspond to training datasets, algorithms, and hyperparameters. The proposal achieves high accuracy when class labels are datasets. Moreover, it can classify algorithms when the training datasets contain the same class. In addition, the discussion highlights that the classification of OCC models is essentially the classification of datasets that treats multiple samples as a single input. The experiment demonstrates the classification of datasets using sleeping records. The proposed method can provide a unified solution for classifying OCC models, datasets, and rankings. Source code is uploaded to the public repository https://github.com/ToshiHayashi/ClassOCC.

Clustering is an unsupervised learning technique to group samples into clusters. Clusteringhas been applied to several data types, but not yet to machine learning (ML)models. The motivation of this paper is clustering ML models. For this purpose, onegeneral idea is to apply k-means. However, averaging ML models is not possible.Accordingly, this paper proposes a novel clustering algorithm, namely k-ensembles.The idea is to apply an ensemble learning to approximate the average of ML models.The proposed method represents ML models as rankings and computes their similarity.The other parts are the same as in k-means clustering. The experiment appliesK-ensembles for clustering one-class classification(OCC) models. The centroids of Kensemblesclustering can act as ensemble models. Moreover, K-ensembles can applyclustering on datasets by treating them as OCC models. Furthermore, K-ensembles canextract feature vectors from ML models using similarities to centroids.

Hyperspectral image classification has made significant progress in recent years with the introduction of the Mamba architecture, which learns long-range dependencies while maintaining linear computational complexity. However, existing Mamba-based hyperspectral image (MambaHSI) classification methods face limitations in handling multi-scale spatial processing and preserving spectral continuity. To address these limitations, this paper proposes a novel MambaHSI classification method, with three enhancements: (1) an enhanced multi-scale spatial Mamba module, which captures multi-scale spatial features while maintaining linear complexity; (2) an enhanced spectral Mamba module, which preserves spectral continuity through cross-group interactions; and (3) an adaptive fusion module, which learns the optimal fusion weights for spatial-spectral features. Extensive experiments on four benchmark datasets demonstrate the effectiveness of the proposed method. Specifically, the method achieves overall accuracies of 97.54%±0.81%, 95.13%±0.83%, 94.37%±0.64%, 96.59%±0.25%, 97.70%±0.64%, and 99.58%±0.23% on the Pavia University, Houston, Hanchuan, Honghu, Longkou, and Botswana datasets, respectively, which represent improvements of 1.80%, 0.67%, 4.16%, 2.01%, 1.21%, and 0.24% over MambaHSI. Comprehensive ablation studies reveal that the multi-scale spatial Mamba module provides the most significant contribution, with improvements of 88.68%, 85.19%, 86.19%, and 92.21% over the baseline on the first four datasets, while the adaptive fusion module provides an additional performance boost of 1.70%∼2.59% across all datasets. The proposed model achieves excellent performance in diverse hyperspectral classification scenarios, while retaining the linear computational complexity advantage of the Mamba architecture.

Achieving collective consensus in large-scale group decision-making (LSGDM) is increasingly challenging due to the explosion of heterogeneous information sources. The core problem lies in distilling coherent outcomes from massive datasets characterized by sparse user ratings, uncertain social trust, and complex strategic interactions. Existing approaches often fragment this problem, treating trust modeling, data completion, and consensus negotiation as isolated processes, thereby failing to capture their intrinsic interdependence. To address this unified challenge, we propose a granular ball (GB) evolutionary game-based framework. Unlike traditional point-based models, we represent each decision maker as a center–radius GB, where the radius inherently encodes uncertainty and social reliability. We introduce an adaptive community detection strategy based on GB distances to reveal latent groups with high consensus potential without requiring prior structural knowledge. Furthermore, we design a two-stage dynamic consensus mechanism that combines payoff-driven intra-group opinion evolution with inter-group coordination. During consensus reaching, individual strategies, opinions, and trust weights interact dynamically, while the previously obtained community structure provides hierarchical guidance for the evolution process. Experimental results demonstrate that the proposed method achieves robust and scalable consensus formation under sparse and uncertain conditions, outperforming existing approaches in adaptability and solution quality. The source code is openly available at https://github.com/JustinaZhan/GBLSGDM.

The Hoist Scheduling Problem (HSP) arises in electroplating lines, where material handling equipment must transport workpieces through sequences of chemical baths within strict time windows. The problem is NP-hard, and its computational complexity renders traditional optimization methods impractical for industrial applications, as solution times typically exceed operational constraints. This paper presents a Deep Reinforcement Learning (DRL) approach to the HSP, employing Deep Q-Networks (DQN) as a proof of concept. The electroplating line is modelled as a Markov Decision Process in which states encode the system configuration, actions represent feasible hoist movements, and rewards quantify scheduling efficiency. Experimental evaluation on a simplified HSP instance demonstrates stable learning dynamics and competitive performance. This work does not aim at outperforming state-of-the-art optimization methods, but rather at assessing the feasibility of framing the Hoist Scheduling Problem as a Markov Decision Process and learning a control policy via Deep Reinforcement Learning. Experimental results on a simplified instance show stable learning behavior and suggest the potential of DRL-based approaches to handle variations in initial conditions and processing times, paving the way for adaptive scheduling strategies in electroplating systems.

The increasing spread of misinformation and ideological polarization in online social networks promotes the formation of isolated and radicalized communities. Understanding how opinions evolve within and across these communities is essential for tracking disinformation, identifying influential users, and analyzing social dynamics. This paper proposes a novel framework for detecting cross-community groups of users who share similar opinions over time. The approach combines a temporal opinion graph based on the Friedkin-Johnsen (FJ) model with a refined Louvain clustering algorithm. The graph representation captures opinion dynamics by integrating user stubbornness, influence relationships, and temporal interactions across connected communities. Building on this model, the proposed clustering method enhances modularity optimization with an opinion-alignment refinement step, allowing clusters to be adjusted according to node-to-cluster opinion similarity. Experiments on real-world Reddit data show that the proposed method improves intra-cluster opinion homogeneity, increases inter-cluster opinion separation, and more effectively identifies cross-community groups of like-minded users compared to standard Louvain clustering. These results suggest that incorporating opinion alignment into community detection better preserves opinion structures in social networks while maintaining competitive modularity.

The SOMET conference (Tools and Techniques of Intelligent Software Methodologies) highlights the state of the art and the new existing trends in software methodologies, tools, and techniques for solving a wide variety of problems present in several branches of both science and technology [...]

Popularity bias has long been a persistent issue in recommendation systems, leading to misleading results and significant problems such as the Matthew Effect and the Information Cocoon Room. Existing studies have primarily focused on the elevation of long-tailed items, overlooking the crucial connection between users and items. Drawing inspiration from causal graphs, this paper introduces a novel framework called CIACC (Causal Inference and Attribute Correlation Consistency) to tackle the challenges posed by popularity bias. The framework leverages causal graphs to evaluate the compatibility between users and items and to gauge the influence of item popularity on rankings. It employs counterfactual inference to estimate the impact of item popularity on rankings and adheres to the consistent principle of attribute correlation to enhance the feature representation of long-tailed items. Through rigorous experiments conducted on three public datasets, we demonstrate that our CIACC framework outperforms state-of-the-art methods.

Thermal infrared (TIR) object tracking is a fundamental task in computer vision that is not affected by changes in lighting conditions. It performs better than visible light trackers in extreme environments such as nighttime, heavy rain, haze, and sandstorms. However, TIR object tracking also faces challenges such as occlusion, thermal crossover, motion blur, and similarity interference. Unlike visual tracking, TIR images lack color information and texture features. These factors make it challenging to learn detailed and shape features of the targets, making it hard to distinguish between targets and interference effectively. In this study, we propose a graph-based deep learning model, SiamMLGR, within the Siamese framework for stable TIR object tracking to address these issues. Specifically, to extract more fine-grained features of TIR targets, we propose a multiple graph attention module (MGAM) to replace the global matching information transmission method in the Siamese framework. This module constructs a graph structure to establish local and global connections between the target and the search area. Furthermore, to retain more of the features learned by the MGAM, we propose a spatial graph convolutional module (SGCM), which uses an explicit graph adjacency matrix to propagate information between the attention graphs. Additionally, we incorporate large-scale datasets from the visual tracking field into the model training process. By mixing these with TIR datasets, we address the sample imbalance issue present in pure TIR datasets. Extensive experimental results indicate that the proposed method achieves state-of-the-art performance.

Magnetic Resonance Imaging (MRI) education often suffers from limited access to physical scanners and the complexity of MRI parameter interdependencies. This paper presents a web-based MRI simulator integrated with a knowledge-based AI assistant to enhance medical and radiography training to bridge the gap between theoretical learning and practical experience in MRI procedures. The simulator offers medical and radiography students an intelligent, interactive platform accessible via any web browser. The simulator enables interactive MRI parameter adjustments and real-time imaging feedback, while the AI assistant employs structured knowledge representation and rule-based reasoning to provide personalized recommendations on parameter optimization, artifact reduction, and protocol selection. The system was developed using modern web technologies including Node.js as a backend solution, JavaScript-driven frontend to scalable with smooth navigation and MongoDB for database. The system’s intelligent assistant leverages domain-specific ontologies and rule-based reasoning to deliver personalized recommendations on parameter optimization, artifact reduction, and protocol selection. The usability testing with 30 medical, system development students and instructors demonstrated high satisfaction, achieving average usability and learning effectiveness scores of 85.1% and 84.7%, respectively. These results indicate the system’s potential to bridge theoretical learning and practical skills in MRI education. Future work will expand anatomical coverage and integrate deep learning to further enhance the AI assistant’s adaptability.

Graph convolutional neural networks (GCNs) have demonstrated effectiveness in processing graph structure. Due to the diversity and complexity of real-world graph data, heterogeneous GCN have attracted significant attention. However, existing research predominantly relies on explicit connections to explore graph heterogeneity. In the case of edgeless graphs, such as information systems, the absence of direct edges poses a significant challenge for employing GCNs to analyze the latent heterogeneity within these graphs. Traditional approaches overlook the topological features of information systems, resulting in information loss. This article introduces a heterogeneous graph convolutional neural network based on shadowed deviation relationship (SDHGCN) to investigate the heterogeneity of information systems, thereby improving the generalizability of heterogeneous GCNs. First, shadow deviation relationship and attribute deviation relationship are constructed derived from shadow sets and information gain, respectively. Then, dexterously integrated with the feature matrix of the information system (the relationship between objects and attributes), a highly expressive heterogeneous graph is constructed. Second, by performing graph convolution operations on the heterogeneous graph, effective node representations can be obtained to complete node classification tasks. Finally, the effectiveness and nonrandomness of SDHGCN are validated by extensive comparison and ablation experiments.

Clustering by fast search and find of density peaks (DPC), a classic density-based algorithm, excels in identifying clusters of arbitrary shape. However, it struggles in recognizing complex structures due to challenges in selecting density peaks and allocating non-central points. To address these issues, we propose an anchor graph based connectivity peaks clustering method which is the connection between anchor graph and density-based clustering, called AG-CPC. Firstly, it introduces a new concept of connectivity by analyzing the divergence and discreteness of neighborhood adjacency graph to detectlow-density clusters and border points. Secondly, a robust two-stage assignment strategy using an adaptive parent-child relationships based on data distribution characteristics, is proposed to reduce the wrong allocation of non-central points. Lastly, a local method for constructing anchor graphs is introduced, combined with fuzzy connectivity and boundary domains of clusters, to scale down the anchor graphs and establish the connection among anchor points. The experiments demonstrate the efficiency and stability of the proposed algorithm compared to state-of-the-art algorithms on synthetic, real-world, and image datasets.

This paper introduces a multi-method explainable artificial intelligence framework designed as a foundational step toward European Union AI Act-compliant automated ICD-10 coding. The framework integrates three complementary explainability methods: label-wise attention, SHAP and case-based reasoning. Unlike existing approaches that typically employ single explanation methods, this new framework creates a comprehensive implementation of multiple explainable techniques while implementing stratified evaluation across different data complexity levels. The architecture leverages a transformer-based model with label-wise attention aggregation. While current performance levels require further development before industry deployment, preliminary results demonstrate competitive performance with micro-F1 score of 0.565 and explanation coverage of 87.1%, establishing critical infrastructure for regulatory-compliant explainable AI in healthcare.

This work presents a new dataset created especially for Visual Question Answering (VQA) on brain tumor MRI images. This dataset includes 750 MRI images of brain tumor with a 512 × 512 pixel resolution. It also includes two different kinds of expert-annotated question-answer combinations in natural language (What/Which, and Yes/No) associated with three possible brain tumor categories (glioma, meningioma, and pituitary). To create a benchmark for this dataset, we propose a dual-stream VQA framework that leverages two transformer-based models to handle image feature extraction, question interpretation, and answer generation. The baseline model is thoroughly assessed on the dataset, revealing the task’s inherent complexity and emphasizing the difficulties in achieving precise medical VQA. The outcomes underscore the dataset’s utility in advancing multimodal medical support systems and lay the groundwork for future progress in this domain.

Large-scale multiattribute group decision making (LS-MAGDM) is common in practice. How to apply it and form a consistent solution has become a hot topic in decision science research. The evaluation information expression and experts' behaviors formed during decision-making process have been shown to have important impact for reaching consensus. Based on this, consensus reaching process for LS-MAGDM considering opinion leaders and two-stage heterogeneous noncooperative behavior is proposed. First, we define cardinal consistency of fuzzy distributed preference relation (FDPR) based on additive consistency. The FDPR score calculation method satisfying additive consistency is presented. Second, we put forward the identification method of opinion leaders. Starting from social network and evaluation information, the concepts of authority, stability, professionalism, and reliability are defined and combined to measure expert's leadership ability. Then, the multiobjective optimization model is constructed to identify opinion leader in each subgroup. Finally, consensus reaching models based on two-stage heterogeneous noncooperative behavior are proposed. noncooperative degree is introduced to measure noncooperative behavior of experts or subgroups. Different feedback mechanisms and noncooperative behavior management methods are applied to both the intra-subgroup and inter-subgroup consensus reaching stages. Finally, the feasibility of this method is verified by a case study of remanufacturer quality assessment and comparative analysis with some State-of-the-Art methods.