Vol. 44 · No. 6

Terahertz imaging is highly effective for detecting concealed objects due to its non-harming nature and its ability to penetrate materials like clothes, paper and plastic, etc. In environment where detection technologies are limited, terahertz imaging emerges as one of the effective and safest methods available. Unlike techniques such as X-rays, it does not emit harmful radiation, making it suitable for surveillance applications. However, many existing object detection models are computationally intensive which can hinder their deployment in real time or resource constrained environments. To address this issue, traditional convolutional operations in the deep learning models have been replaced with depth-wise convolutions and depth-wise separable convolutions in proposed approach. These modifications significantly reduce the number of trainable parameters and computational load during model training. The optimized architecture has been integrated into widely used object detection models – namely YOLOv5m and YOLOv8m, using terahertz images of concealed objects as input. This integration enhances training efficiency with minimal loss in accuracy, making the models more suitable for deployment on devices with limited computational power and memory.

In a complex network comprising many devices, a set of nodes may be partitioned into multiple local clusters with distinct functions, properties, or communication protocols. Thus, there has been an increase in network design problems with additional constraints regarding the clustering of vertices, one of which is the Clustered Steiner Tree Problem – a variant of the Steiner Tree Problem. There have been a few studies working on this problem in the literature, but they either solve it only in the metric case or their exploration capability remains limited. Therefore, their results are not good in many cases. To overcome the drawbacks, we propose a Priority-Based Genetic Algorithm to solve the Clustered Steiner Tree Problem. The proposed algorithm maintains a balance between exploration and exploitation to prevent the search from getting stuck in local optima. Experiments and comparisons to existing works in non-metric and metric cases are carefully conducted to prove the remarkable performance of the proposed algorithm.

With the increasingly widespread application of MAS (Multi-Agent System), recent years have witnessed growing research interest in the verification of MAS properties. Similar to traditional concurrent and multi-component systems, the paramount challenge in MAS verification is the state space explosion problem. One approach to mitigating this issue is to reduce the original model before generating the state space, as reductions at the model level can often substantially alleviate state space complexity. However, for MAS modelled using Petri nets, existing methods primarily focus on reduction at the state space level. In contrast, this paper addresses reduction at the underlying model level by proposing a solution of structural reduction for Petri nets that preserves temporal epistemic logic properties. The main contributions consist of three aspects. Firstly, since existing structural reduction rules for Petri nets are not suitable for temporal epistemic logic verification, modifications and extensions to some of the rules are introduced to accommodate temporal epistemic logic, and corresponding theorems are provided to guarantee the correctness of these rules. Furthermore, given that the applicability of structural reduction rules in Petri nets is constrained by transition visibility, this paper conducts a refined visibility analysis of transitions based on semantic characteristics of epistemic logic for a subclass of Petri nets and a specific category of properties. The resulting analysis relaxes the visibility constraints when applying the rules, and its correctness is formally guaranteed by a theorem. Finally, case studies demonstrate that the proposed structural reduction rules save space overhead during verification and achieve better reduction performance under relaxed visibility constraints.

With the growing demand for personalised travel experiences, the development and application of travel point-of-interest (POI) recommendation systems have become increasingly important. However, many existing systems often underperform owing to insufficient integration of social features and contextual information. To address this issue, the S-LMF algorithm is proposed, combining social features with logistic matrix factorisation to improve recommendation accuracy. This approach simulates social influence by incorporating joint check-in similarity and friendship factors, while logistic matrix factorisation leverages check-in frequency data to refine POI recommendations. The effectiveness of social features and logistic matrix factorisation (S-LMF) was tested against five baseline algorithms using publicly available data sets from Yelp and Gowalla. Results demonstrated that S-LMF outperformed the best baseline model by improving @20 by 22.95

Business processes realized by enterprise information systems are usually designed and verified in advance by process models. Event logs generated from such systems can be used to ensure the correctness of the business process. With the upgrade of systems or changing of customers' requests, it often appears that the actual behaviors observed in event logs are not consistent with those in process models. Process mining techniques are thus used to construct and repair process models by mining event logs. In this work, we propose a new model repair approach based on logic Petri nets (LPNs). It can repair a Petri-net-based process model containing a loop return structure. According to alignments, relations between the activities in event logs and the process models are analyzed such that the deviations can be found. Then, logic expressions of logic transitions are constructed and an LPN-based process model is thus constructed. The repaired model can accurately describe the same processes as the event logs. Finally, some cases related to a thoracic surgery process in a hospital are given. The correctness and effectiveness of the proposed approach are illustrated by experiments.

This article focuses on the design and implementation of a 3D Convolutional Neural Network (3D-CNN) for hyperspectral classification of tree species. Real data from repeated aerial imaging of selected sections of the Slovak electricity transmission system was used to test the models, using technology from our partner VUJE, a.s., which has a hyperspectral scanner consisting of two cameras, HySpex VNIR-1800 and HySpex SWIR-384, capturing wavelengths from 400 to 2 500 nm. The results of 3D-CNN classification based on autumn and spring data collection were compared, as well as classification after fusing data from selected areas for the purpose of comparing object vs. pixel classification. The presented object-based classification model based on 3D-CNN achieves on average 9

Transformer models and their variants have shown significant advantages in natural language processing tasks, but their high computational requirements limit their deployment on resource-constrained devices. To achieve the balance of computational cost and accuracy, the lightweight model pNLP-Mixer utilizes parameter-free projection to generate text embeddings. Nevertheless, the text embeddings generated by projection involve shallow semantic information and ignore the exploration of implicit semantic information. To overcome the challenges of high parameter cost and insufficient representation capabilities in existing models, we propose a method that incorporates topic information to enhance semantic richness. Our approach leverages the Latent Dirichlet Allocation (LDA) topic model to capture latent semantic relationships between words, thereby improving the expressiveness of text representations for downstream tasks. Building on this method, we propose a lightweight model named TEP-Mixer, which integrates multiple feature extraction modules to further enhance representation capabilities. Experimental results demonstrate that TEP-Mixer outperforms other lightweight models in accuracy while maintaining a lower parameter count across multiple benchmark datasets. It is suitable for resource-constrained devices.

Knowledge tracing endeavors to track shifts in students' knowledge states throughout the learning process by utilizing students' work records, enabling the prediction of their responses to the next question. As knowledge tracing advances, more information can be mined from the datasets to improve the model. However, most existing models only consider the correctness of the question and the concept as input, ignoring much of the other information gathered by intelligent tutoring systems (ITS). Therefore, this paper presents a Behavior-Enhanced Knowledge Tracing Model (BEKT). The Behavior-Enhanced Knowledge Tracing Model first selects multi-dimensional features by calculating the mutual information. The mutual information value shows that the most relevant features to the prediction target are student behavior information, including the count of hints, attempts made, and time taken to answer questions. Then, the model proposes a cross-feature fusion method to change the students' knowledge mastery state. Finally, a multi-layer perceptron is applied to integrate student behavior features to enhance knowledge state representation. Compared to existing models, BEKT more accurately captures changes in knowledge state, improving model performance.

The Manufacturing Execution System (MES) is a critical component of intelligent manufacturing systems, enabling the automation and intelligent control of production processes. However, conventional MESs generally lack sufficient flexibility in coping with process reconfiguration and dynamic resource variations, thereby resulting in production bottlenecks and prolonged manufacturing time. To alleviate production bottlenecks while reducing overall manufacturing time, we propose a modeling, analysis, and resource allocation optimization framework for MESs. This framework employs rewritable timed Petri nets (RTPNs) to model and analyze MES behavior. Furthermore, a resource allocation optimization algorithm is developed to minimize production time. A clothes customization manufacturing system is adopted as a case study to demonstrate the effectiveness of the proposed method. The production process is reconstructed and optimized based on the RTPN model, and system performance is validated through simulation. Experimental results indicate that the proposed method significantly reduces production blocking and waiting rates, thereby improving overall operational efficiency.