Maximum Sustainable Yield (MSY) and Maximum Economic Yield (MEY) are key concepts in population and resource management, critical for optimizing exploitation and ensuring long-term sustainability. MSY represents the largest possible yield that maintains the resources’ ability to regenerate, while MEY maximizes the economic benefits. Traditional models, such as Schaefer’s logistic growth, provide valuable insights but often oversimplify ecological dynamics, social influences, and spatial factors. In contrast, Agent-Based Modeling (ABM) offers a dynamic and flexible approach by simulating complex interactions between agents and their environment and incorporating stochastic events, thus better capturing real-world complexities. This paper explores the potential of ABM to enhance the understanding of MSY and MEY in fisheries management, using the case of anchovy fisheries in the Gulf of Thailand as a study. Results from simulations demonstrate ABM’s effectiveness in analyzing both biological and economic outcomes, revealing differences between traditional and ABM-derived estimates of MSY and MEY. While still in the early stages of application, ABM shows promising results for fisheries management, making it valuable for both scientists and stakeholders. The research may also open various developing and applicating opportunities in similar fields. However, it also suffers from some notable drawbacks in the development stage such as intensive requirements of time and resources, thorough consideration of the model’s realism trade-off, and lack of initial input.
Historical menus serve as valuable archival records that reflect culinary practices, consumption patterns, and environmental changes. However, limited research has explored their potential in understanding shifts in culinary trends. This study addresses this gap by investigating the effectiveness of word embeddings and optimization techniques in classifying cuisine patterns using historical menu data. By comparing traditional, ensemble, and deep learning models, the paper focuses on using word embeddings for multi-class classification. TF-IDF embeddings consistently provided superior performance, especially in traditional models like Support Vector Machines and ensemble methods such as Random Forest, achieving the highest accuracy, 0.9059, and F _1 -score, 0.9011, when combined with Word2Vec. Findings indicate that selection of mere embedding does not effect the model’s performance for this classification task, hybrid embeddings with hyperparameter optimization improve classification performance. Future research should investigate transformer-based models to better capture ingredient and cuisine nuances.
We propose a novel method for image gradient upsampling that leverages adaptive convolution with learnable kernels to add weighted permutations of the input image pixels back to the image in a constrained manner in order to enhance the image gradient resolution. The proposed approach dynamically adjusts pixel values based on local patterns of pixels in the input image that define the gradients in the image. Traditional upsampling techniques, such as unpooling and nearest-neighbor interpolation rely on predefined rules and often result in blurred or jagged edges due to their inability to adapt to image content. Moreover, these and other image upsampling techniques like those based on deep learning, focus on enhancing the spatial information in the input with the goal of enhancing the human visual perceptual experience. However, the goal of our approach is to enhance the gradient information in the image in order to help gradient based learners learn better latent space features for inference tasks. Our approach uses trainable convolutional kernels that learn to extract and amplify intricate gradient based features during training by subdividing the input images into overlapping patches and applying element-wise transformations. We show the efficacy of the proposed method by enhancing the gradient information in images of the Fashion MNIST dataset.
This work addresses the facility location problem, a key area of research in Operations Research and Artificial Intelligence. Specifically, we examine a competitive facility location problem where a firm seeks to establish new facilities in a market already served by existing competitors. To predict customer demand, we utilize a general class of customer behavior models, known as the nested-logit model, which is widely recognized as one of the most popular demand models in the literature. The facility location problem under the nested-logit model is characterized by its high nonlinearity and complexity. Existing methods either do not operate within polynomial time or fail to guarantee near-optimal solutions at any desired level of precision. In this study, by leveraging the unique structure of the nested logit choice model, we propose a Fully Polynomial-Time Approximation Scheme (FPTAS) to efficiently solve the problem. To the best of our knowledge, this is the first FPTAS developed for this type of competitive facility location problem.
This research presents a Buddhist tourism recommendation system aimed at enhancing personalized travel experiences through image classification and tourist behavior analysis. Focusing on Buddhist landmarks in Thailand, the system uses Convolutional Neural Networks (CNNs) to classify and categorize key features of Buddhist sites, such as architectural styles and religious symbols. Recommendations are generated based on tourist preferences, gathered from sources such as social media and surveys, which reveal patterns in spiritual, historical, and cultural interests. The system architecture includes a Data Collection Layer for gathering images and user interactions, a Data Preprocessing Layer for cleaning and augmenting the data, and a Classification Layer for feature extraction. The AI Engine Layer employs hybrid filtering methods, combining collaborative and content-based filtering to provide context-aware recommendations. A Personalization Module in the Application Layer customizes the user interface to enhance engagement. The system aims to distribute tourist traffic more evenly, promote lesser-known sites, and support cultural preservation by offering tailored experiences. The dataset for image classification consists of 1,000 images of Buddhist Phra That, categorized into 11 classes, sourced from travel blogs and social media. The dataset was split into 80
The field of legal assistance systems has gained significant attention for its benefits, such as increased efficiency in legal case processing, reduced workload of legal practitioners, and improved accessibility. While there are several developments in this domain, only a handful of inventions address the inheritance calculation, and to the best of our knowledge, none directly handle the intestate inheritance allocation per the Thai Civil and Commercial Code. Therefore, we developed a novel algorithm to automate such a process utilizing the rule-based algorithm and tree-based algorithm, which include the recursive non-binary tree construction, bottom-up pruning, and result allocation. This algorithm can correctly allocate inheritance for all test cases, indicating its potential for adoption in practical applications in the future.
Misinformation prevention is a crucial research topic in the social network community, which aims to identify, mitigate, and ultimately prevent the spread of false or misleading information across the social network. Traditional methods typically rely on detailed information on underlying networks and propagation mechanisms. However, these approaches can be impractical due to privacy concerns and restrictions on data accessibility. To overcome this challenge, we propose MetaLearner, a novel framework that leverages historical pairs containing queries and their high-quality decisions. This approach enables us to derive effective decisions for new queries without requiring complete network and diffusion information. We evaluate the performance of MetaLearner by comparing it with existing methods. The results show significant improvements in both accuracy and computational efficiency, demonstrating MetaLearner as a scalable and effective solution for misinformation prevention.
This paper discusses the creation and use of the numerical model for flood assessment in the Nhat Le river basin, Vietnam. In particular, the hydrodynamic model based on Mike Flood has been calibrated and validated with the observed water levels at gauging stations, flood marks, and flood extent derived from Sentinel 1 satellite imagery in the historical flood events of October 2010 and October 2020. The well-validated model was adopted to determine the flood damage risk for the structures and relevant elements located in the basin. The developed mathematical model allows one to assess the flood risk in the Nhat Le river basin, as well as the implementing and planning for individual recovery measures and general shelter. It can also be used to help make decisions like any combination of land use, land-cover types, socio-economic activities, building patterns, and flood control strategies.
Content-based image retrieval (CBIR) with a static threshold often encounters limitations due to the varying characteristics of queries in different image galleries. In this paper, we propose an approach to address this challenge by introducing dynamic threshold determination for image retrieval. Our method dynamically adjusts the threshold for each gallery based on the distribution of galleries close to that gallery in feature space. By tailoring the threshold to the specific characteristics of each gallery, our approach aims to enhance retrieval accuracy and relevance. We evaluate our method using the ROxford dataset and compare it with the best static threshold. Our approach yields significant enhancements in macro F1 scores across diverse dataset complexities. In the ROxford (medium) scenario, we observed an 12.4
Multiscale random Bernstein polynomial (msBP) priors exhibit many favorable properties for nonparametric Bayesian inference. In msBP, an infinite tree of probability weights is generated from a generalization of the stick-breaking process representation of the Dirichlet process, with each tree scale featuring a weighted random Bernstein polynomial. The prior can generate densities with locally varying smoothness, accommodating abrupt local changes. The degree of smoothness of the resulting function approximation is determined by a hyperparameter that controls the decline in probabilities over the scales. In this paper we extend the multiscale Bernstein polynomial prior from [0, 1] analyzed in the previous literature to the unit hypercube [0,1]^d for the link function in copula dependence modeling, accounting for the uncertainty inherent in the degree of approximation smoothness by Bayesian Model Averaging over the smoothness hyperparameter. The favorable properties of the resulting non-parametric copula model are attained at the cost of an increased computational burden in practical implementation. We provide details of the implementation algorithm that is based on large-scale parallelization tailored to the internal model structure. We implement the algorithm on distributed nodes of a Unix High Performance Computing system with Graphical Processing Units (GPU) acceleration via a Message Passing Interface (MPI) in Modern Fortran. The algorithm scales well across MPI ranks. The routines can be pre-compiled into numerical libraries and invoked from high-level languages such as Python or R.
Prevention of cybersecurity risks in contemporary communication systems depends on phishing email detection. This work presents an optimal deep learning method using the Hill Climbing (HC) algorithm for hyperparameter optimization and BERT for feature extraction to improve phishing detection. Using a Kaggle dataset, the model was trained with balanced precision, recall, and F1-scores for phishing and safe emails with a 95
This research aims to explore the key features influencing the number of respiratory disease cases in Thailand. We analyzed various features to identify their correlation with respiratory disease rates per 100,000 people. By utilizing clustering techniques, we grouped provinces with similar characteristics to better visualize the relationships between these features and respiratory health outcomes. This method provides insights into the types of urban and environmental characteristics that contribute to the incidence of respiratory diseases. Our findings reveal that air quality, influenced by urban expansion and agricultural activities, is the most significant feature associated with high respiratory disease rates. Following this are severe environmental conditions, while local environmental features and employment practices shaped by provincial development policies also play a crucial role.
The future of transportation is rapidly transforming with the increasing adoption of electric vehicles (EVs), signifying a global commitment to sustainability. This expansion introduces significant challenges, particularly in the construction and maintenance of EV charging infrastructure. As EV numbers rise, user dissatisfaction due to inadequate charging facilities threatens the adoption of electric mobility. Additionally, the strain on financial and power resources necessitates strategic planning to avoid misallocation. This paper focuses on addressing these challenges within the context of charging stations inside a university campus where these challenges are significantly emerging due to the synchrony of EV users (arrival time, charging time,...) as well as electrical uses with different purposes (charging and other activities) at the same time. To tackle these issues, we propose a Simulation Environment using an Agent-Based Modelling Approach (ABM). We simulate the charging behaviours of EV users, optimize infrastructure deployment, and enhance decision-making for charging station operations thanks to simulations based on real data. The results showed that increasing active charging ports in targeted areas, combined with strict rules banning gasoline vehicles from EV spots, achieved over 90
This research focuses on optimizing escape routes in emergency scenarios involving active threats to minimize risks to lives and property. We evaluate the effectiveness of four algorithms Ant Colony Optimization (ACO), Bellman-Ford, Dijkstra, and A* in determining the most efficient escape routes across 50 maps. The primary objective is to identify paths that not only provide the shortest escape route but also reduce potential exposure to the threat. Our results demonstrate that the ACO algorithm consistently delivers the best performance, offering optimized escape routes that meet these criteria more effectively than the other algorithms. By prioritizing safety and minimizing exposure, this approach contributes valuable insights to emergency response planning in high-risk situations, potentially mitigating harm to individuals and property. The comparison and analysis of these algorithms provide a foundation for future advancements in intelligent navigation systems for crisis management.
The transition from a linear to a circular economy (CE) emphasizes sustainable production and consumption by promoting resource efficiency, reuse, and recycling. This systematic literature review explores how artificial intelligence (AI) technologies, particularly machine learning (ML), can enhance various stages of CE, including raw material extraction, product design, manufacturing, distribution, and waste management. A total of 14 peer-reviewed studies published between 2020 and 2024 were analyzed, identifying key applications and trends in AI-driven CE practices. Findings highlight that AI optimizes processes such as sustainable concrete design, CO2 tracking across supply chains, smart water distribution networks, and automated waste sorting. Although AI presents significant opportunities to increase resource efficiency and reduce waste, challenges remain, such as high computational requirements and the need for quality data. This review underscores the potential of AI as a critical enabler of circular strategies, while also pointing to the importance of further research to address current limitations.
We propose a new graph-based method for exploration and representative set selection in vector spaces for classification tasks, and demonstrate the results on a well-known MNIST dataset. Our method reveals significant variations in the similarity structure among vectors corresponding to different label classes. To better capture these differences, we construct similarity-based graphs (networks) with a balanced number of edges within each label class, rather than using a single similarity cutoff across all classes. This tailored approach provides deeper insights into class-specific similarities. We further analyze the resulting graphs by computing various global and local structural properties. Additionally, we introduce a representative set selection model based on independent and dominating sets. We validate the effectiveness of our approach by training a Convolutional Neural Network (CNN) on these representative sets and comparing performance against randomly selected sets of images of the same size.
Skin lesion classification is a significant challenge in the medical field, particularly in the early detection of malignant skin lesions. Traditional methods require large amounts of data and often struggle to handle heterogeneous datasets, especially in medical datasets where there is an uneven distribution of disease types. A novel approach has been proposed that integrates Energy-Based Models (EBM) with an Energy Distance (ED) approach. This method aims to incorporate ED into the training process of the EBM model. The experimental results on the ISIC_2019 dataset achieved classification accuracy of 74.19
Medical Visual Question Answering (VQA) has become increasingly significant in aiding physicians with disease diagnosis and providing patients with detailed insights into their conditions. However, Medical VQA still lags behind general VQA due to challenges such as limited availability of accurate data and the complexity of medical terminology. Existing models frequently struggle with performance issues caused by the intricate nature of image and text encoders. Recent research has concentrated on improving fusion modules for integrating question and image features and employing pre-trained models with self-collected datasets, but often overlooks the value of question and image history. This paper proposes an approach that introduces an Associative Memory block to leverage historical questions and images, enhancing the vision-language context. Additionally, we incorporate a Prototype Learning block that utilizes hierarchical prototype learning on text and image embeddings through advanced Hopfield layers. Our approach focuses on identifying the most representative prototypes from text-image embeddings, enriched by associative memory, rather than learning direct text-image joint feature representations. This method facilitates a more nuanced representation of semantics for answering questions. Our proposed approach achieves the best performance on the VQA-RAD dataset, demonstrating a significant accuracy improvement of 0.45
Legal natural language processing has recently received a surge in interest from experts due to its potential application in various domains. One of the most popular tasks in this area is legal case categorization and judgment prediction, which can be seen as a classification problem. Previous attempts at the classification task mostly employ traditional methods, such as the static embedding method in combination with machine learning or deep learning models. Even though such approaches can achieve a notable performance, they have yet to reach state-of-the-art performance. With the significant leap in text and natural language processing, namely the advent of transformer-based models, along with their consistent improvement, we see the opportunity to adopt this development to the legal classification task. Our proposed modeling pipeline, the WangchanBERTa-SFT-sc, can outperform the baseline model using the fastText embedding method and conventional machine learning models on the binary classification of whether or not the case is related to property offence, reaching exceptional test accuracy and F1-score of 94.5
Generative AI emerged as a powerful method for generating new data, which is famous for image, video, and text generation. For this inspiration, our study aims to leverage the Generative AI for Network Traffic Dataset Generation. This type of data is very important to cybersecurity, as it is used for detection, prediction, and classification tasks in the network traffic. Recent studies have utilized the variant of Generative AI models for generating Network Traffic Data, gaining considerable achievements. In our study, we proposed a novel approach using the One-Dimensional Generative Diffusion Model to generate the synthetic Network Traffic data. As a result, our proposed model surpasses the previous studies’ models on the UNSW-NB15 dataset, with the Chi-Squared Test and Kolmogorov-Smirnov Test values being 0.863 and 0.838 respectively. Moreover, our proposed method considerably improves the performance of detecting attack activities of network traffic in terms of accuracy and F1-score, demonstrating its powers in cybersecurity.
