This study investigates embedding reconstruction attacks in large language models (LLMs) applied to genomic sequences, with a specific focus on how fine-tuning affects vulnerability to these attacks. Building upon Pan et al.'s seminal work demonstrating that embeddings from pretrained language models can leak sensitive information, we conduct a comprehensive analysis using the HS3D genomic dataset to determine whether task-specific optimization strengthens or weakens privacy protections. Our research extends Pan et al.'s work in three significant dimensions. First, we apply their reconstruction attack pipeline to pretrained and fine-tuned model embeddings, addressing a critical gap in their methodology that did not specify embedding types. Second, we implement specialized tokenization mechanisms tailored specifically for DNA sequences, enhancing the model's ability to process genomic data, as these models are pretrained on natural language and not DNA. Third, we perform a detailed comparative analysis examining position-specific, nucleotide-type, and privacy changes between pretrained and fine-tuned embeddings. We assess embeddings vulnerabilities across different types and dimensions, providing deeper insights into how task adaptation shifts privacy risks throughout genomic sequences. Our findings show a clear distinction in reconstruction vulnerability between pretrained and fine-tuned embeddings. Notably, fine-tuning strengthens resistance to reconstruction attacks in multiple architectures—XLNet (+19.8%), GPT-2 (+9.8%), and BERT (+7.8%)—pointing to task-specific optimization as a potential privacy enhancement mechanism. These results highlight the need for advanced protective mechanisms for language models processing sensitive genomic data, while highlighting fine-tuning as a potential privacy-enhancing technique worth further exploration.

The rise of new disease variants, such as COVID-19, influenza, and others, highlights the critical need for advanced epidemiological modeling to guide early-stage outbreak management, especially when vaccine options are not available or reliable. This paper presents a novel, hybrid, data-driven model that integrates Agent-Based Modeling (ABM) with an extended SEIHRD (Susceptible, Exposed, Infectious, Hospitalized, Recovered, and Dead) framework, enhanced by N-step Deep Q Reinforcement Learning (N-Step DQRL). This model merges ABM’s behavioral insights with the SEIHRD model’s progression dynamics, utilizing DQRL for adaptive, data-informed decision-making. It is particularly focused on enhancing non-pharmaceutical interventions, such as lockdown policies, which are crucial in managing outbreaks in the absence of vaccines. This approach strikes a balance between detailed analysis and scalability, vital for policymakers in responding to emerging disease variants. The model’s efficacy, as evidenced by an analysis of recent COVID-19 data, highlights its potential to significantly improve global pandemic preparedness and response, merging behavioral analysis with disease progression trends through the use of advanced deep learning techniques.

A biomarker identification model, integrating natural language processing (NLP) and graph convolutional neural network (GCN), offers a novel approach to enhance a simple neural network’s ability to capture the contextual semantics of genes and extract spatial feature information by utilizing gene ontology (GO) annotations. First, we explore gene expression datasets to identify differentially expressed genes (DEGs) and construct a protein-protein interaction (PPI) network. By employing Word2Vec, an NLP algorithm, for vectorizing GO annotations, our model reveals complex biological relationships among genes. GO annotations are crucial as they provide comprehensive information about gene functions, biological processes, and cellular components, thus augmenting our understanding of how genes interact within the network. Integrating multi-layered GCN facilitates effective learning of complex semantic relations and spatial feature information within the PPI network. Experiments on publicly available datasets of Glioblastoma Multiforme (GBM), the most aggressive form of brain tumour, demonstrate that our model significantly enhances biomarker identification compared to existing state-of-the-art methods, showcasing its potential for advancing GBM research and clinical decision-making.

The diversity and complexity of cancer pose significant challenges in creating target treatment strategies. Identifying molecular subtypes of cancer is crucial for recognizing patients with distinct molecular profiles, thereby enhancing the accuracy of diagnosis, prognosis, and treatment decisions. With recent advancements in technology, there’s been a significant increase in the availability of multi-omics data, which is instrumental in the understanding of different cancer subtypes. However, accurately subtyping cancer is difficult due to the high dimensionality and heterogeneity of omics data. Current research in subtype identification often consolidates multi-omics data into a single dataset through simple concatenation and then employs machine learning models to derive a lower-dimensional representation, neglecting the unique distributions of different omics data types. Additionally, they separate representation learning and clustering into two stages, initially learning latent representations and then applying clustering algorithms, leading to suboptimal results due to overlooking the intrinsic clustering structures in the initial learning phase. To address these limitations, we propose a novel deep unsupervised learning model, Subtype-MMCC (Multi-modal Contrastive Clustering) that combines a multi-modal architecture with decoupled contrastive clustering to create an end-to-end framework. Tested on eight TCGA cancer datasets, Subtype-MMCC outperforms existing clustering methods, with its efficacy further validated by survival and clinical analysis outcomes.

In the dynamic landscape of social network analysis, the accurate prediction of implicit relationships presents a pivotal challenge. This paper introduces an innovative solution, the Relation Temporal Graph Convolutional Network with Confidence (R-CTGCN), specifically designed to address the intricate task of predicting implicit relations within evolving social networks. R-CTGCN unifies timestamp temporal embeddings, confidence metrics, and PFs within a comprehensive graph neural network framework, aiming to capture the evolving dynamics of networks and enhance predictive accuracy. Experimental evaluations conducted on diverse datasets, including Epinions and Enron, showcase R-CTGCN’s superior performance compared to both baseline models and contemporary state-of-the-art methods. The emphasis on the roles of confidence and PFs underscores their significance in implicit relationship prediction. The outcomes contribute substantively to the understanding of predicting implicit relationships, positioning R-CTGCN as a robust tool tailored for complex social network scenarios.

Anomaly detection in computer vision-based quality control systems is crucial for industrial defect identification. This research presents the Many-to-One (M2O) framework, employing a multi-level transformer encoder and a single transformer decoder to detect and localize anomalies. With the advent of Industry 4.0 and electric vehicles, this area has gained significance. Despite prior contributions, challenges in generalization and time complexity persist. The M2O framework addresses these issues, enhancing robustness and efficiency in anomaly detection. M2O employs a transformer-based architecture and introduces the Multi-Level Feature Fuse module. To establish a benchmark for industrial electrical connectors, the ECAD dataset, containing real-world anomalies, is introduced. This dataset can inspire further research. Through evaluation against MVtec AD, BTAD, and ECAD, M2O demonstrates superior performance, overcoming previous limitations and offering a robust solution for industrial anomaly detection.

The importance of understanding temporal dynamics in accurate recommendation systems is widely acknowledged. Sequential recommendation systems can efficiently model the dynamics of users and items over the period of time. Therefore, considering long-term and short-term interests of the user is essential for accurate recommendation. However, existing models considered users' long-term and short-term behavioural patterns, but ignore the side information, which plays an important role in improving the performance of the recommendations. As user behaviors contain a lot of information, such as consumption habits and dynamic preferences. In order to better locate user interests, our model called XLNet4Rec, considers the side information (additional features) along with IDs as inputs. Apart from this, unidirectional architectures such as GRU, LSTM restrict the power of hidden representation of users' behavior sequences and they follow the rigid ordered sequence, which is not always true in real world applications. Therefore, we used Transformers4rec (end-to-end RecSys framework), which consists XLNet, a transformer based architecture, employs the deep bidirectional approach to model user behavior sequences and also efficient in processing multiple features. In this paper, we proposed a recommendation system based on users' long-term and short-term interest using XLNet to predict the next user item interaction. Our model improves the quality and personalization of item recommendations for users. In this paper, we conducted experiments on two real-world datasets: Movielens and REES46. Empirical results show that our proposed model is more effective in recommending relevant items to the user compared to previous approaches.

In order to complete a common goal, team recommendation problems identify an efficient group of experts who can collectively satisfy a set of required skills. A significant number of studies address this problem through graph-based approaches. Recently, researchers have started to see this problem as a social information retrieval and examine it through neural architectures that recommend the team of experts by learning a relationship between the skills and experts space. However, this learning process faces several challenges including (1) being unable to handle the modification of a network if the training process is over, (2) the time complexity of the learning process being high and proportional to the size of the network. In this paper, we propose a new architecture, LANT, which comprises transfer learning and neural team recommendation, to address these challenges based on graph neural networks and variational inference. Since the transfer learning of team recommendation is an unsupervised task, therefore, to learn node embedding in a self-supervised manner, we use Deep Graph Infomax with Graph Attention Networks as an encoder. We empirically demonstrate how LANT overcomes the challenges in the existing approaches and compare them against the state-of-the-art approaches in terms of effectiveness using the DBLP dataset.

In this research, we explore the potential of combining effective feature extraction techniques with traditional machine-learning algorithms to classify different types of pneumonia from chest X-ray images. The accurate identification of COVID-19 pneumonia, as well as differentiating it from normal X-rays and other viral pneumonia cases, is crucial in supporting physicians with efficient and reliable diagnoses during times of heavy pressure on the medical system. We present a machine learning-based model for classifying COVID-19 pneumonia-affected lungs, non-COVID pneumonia-affected lungs, and healthy lungs based on chest X-ray images. Specifically, we employ a local binary pattern (LBP) feature extraction technique in conjunction with a support vector machine (SVM) model to achieve 100% accuracy in distinguishing COVID-19 pneumonia from normal X-ray images, which outperforms state-of-the-art methods. In the multiclass classification task, we utilize a Histogram of Oriented Gradients (HOG) feature extraction technique with an SVM-based model, attaining an accuracy of 94%. These results are highly competitive with the performance of deep learning methods commonly used in this domain.

As the world recovers from the COVID-19 pandemic, there is a growing need for effective strategies to prepare for future health crises. Artificial Intelligence (AI), driven by comprehensive and up-to-date data, can play a crucial role in addressing such challenges. Focusing on Canadian data, this study demonstrates the importance of extensive data collection and its implications for global health crisis management. Using feature extraction and deep learning-based regression techniques, we identified key predictors of COVID-19, achieving an$R^{2}$of 0.93 and 0.80 for predicting new cases and deaths. The results emphasize AI's potential in guiding data-driven strategies, stressing the need for global collaboration in data collection and AI deployment to prepare for future health crises.

Social isolation is a serious public health issue that can lead to various mental and physical health problems for individuals and jeopardizes their life's quality. The issue is more critical for older adults and palliative patients who are already suffering from different diseases and lack some abilities for performing their daily tasks. Additionally, this situation worsens when the COVID-19 pandemic adds forced social isolation to people's lives worldwide. In this paper, we propose a framework for detecting social isolation in community-based palliative care networks. We look at the problem as an outlier detection in community-based social graphs. Hence, we map the network to an attributed weighted social graph. Consequently, each patient is linked to a set of informal and formal care providers. We define formulae and indices to extract the norm of the society in terms of structural connections and assign a value to each individual based on the quality and quantity of its connections. The structural indices and a set of quality of life features such as age, marital status, life satisfaction, and capabilities are then used to identify the isolated individuals. We analyze and evaluate the performance of our algorithm on real-life data obtained by the Windsor Essex Compassion Care Community (WECCC), as well as various synthetic social graphs.

Abstract Literature have focused on studying the apparent and latent interactions within social graphs as an n‐ary operation, which yields binary outputs comprising positives (friends, likes, etc.) and negatives (foes, dislikes, etc.). Inasmuch as interactions constitute the bedrock of any given social network (SN) structure; there exist scenarios where an interaction, which was once considered a positive, transmutes into a negative as a result of one or more indicators which have affected the interaction quality. At present, this transmutation has to be manually executed by the affected actors in the SN. These manual transmutations can be quite inefficient, ineffective, and a mishap might have been incurred by the constituent actors and the SN structure prior to a resolution. Our problem statement aims at automatically flagging positive ties that should be considered for breakups or rifts (negative‐tie state), as they tend to pose potential threats to actors and the SN. Therefore, we have proposed ClasReg: a unique framework capable of breakup and link predictions.

Knowledge Graph represents the real world's information in the form of triplets (head, relation, and tail). However, most Knowledge Graphs are highly incomplete. The goal of a Knowledge-Graph Completion task is to predict missing links in a given Knowledge Graph. Various approaches exist to predict a missing link in a Knowledge Graph, but the most prominent approaches are based on tensor factorization and Knowledge-Graph embeddings, such as RotatE and SimplE. The RotatE model depicts each relation as a rotation from the source entity (Head) to the target entity (Tail) via a complex vector space. In RotatE, the head and tail entities are derived from one embedding-generation class, resulting in a relatively low prediction score. SimplE is primarily based on a Canonical Polyadic (CP) decomposition. SimplE enhances the CP approach by adding the inverse relation where head embedding and tail embedding are taken from the different embedding-generation class, but they are still dependent on each other. However, SimplE is not able to predict composition patterns. This paper presents a new, hybridized variant (HRotatE) of the existent RotatE approach. Essentially, HRotatE is hybridized from RotatE and SimplE. We have used the principle of inverse embedding (from the SimplE model) in a bid to improve the prediction scores of HRotatE. Hence, our results have proven to be better than the native RotatE. Also, HRotatE outperforms several state-of-the-art models on different datasets. Conclusively, our proposed approach (HRotatE) is relatively efficient such that it utilizes half the number of training steps required by RotatE, and it generates approximately the same result as RotatE.

Coronavirus disease 2019 (COVID-19) has been an ongoing threat to the world's health system. Millions of people died all over the world because of this deadly virus outbreak. Although health sectors are equipped with modern technologies yet struggling every day to control this outbreak. However, predicting the active COVID-19 cases and hospitalization in advance can be helpful to minimize the catastrophe of this persistent outbreak. This study proposed a novel Agent-based modelling (ABM) framework based on various temporal and non-pharmaceuticals parameters to predict active cases and hospitalization cases. We evaluated the model's performance based on COVID-19 data of Windsor-Essex county region of Ontario, Canada, and eventually achieved satisfactory results in predicting active cases and hospitalization. Experimental results have demonstrated that the simulations provide helpful information that could help take advanced steps to cover up for the shortage in hospital resources and take necessary steps to reduce the number of infections.

The proliferation of fake news on social media sites is a serious problem with documented negative impacts on individuals and organizations. A fake news item is usually created by manipulating photos, text, or videos that indicate the need for multimodal detection. Researchers are building detection algorithms with an aim for high accuracy as this will have a massive impact on the prevailing social and political issues. A shortcoming of existing strategies for identifying fake news is their inability to learn a feature representation of multimodal (textual+visual) information. In this paper, we present a novel approach using a Cultural Algorithm with situational and normative knowledge to detect fake news using both text and images. An extensive set of experiments have been carried out on real-world multimedia datasets collected from Weibo and Twitter. The proposed method outperforms the state-of-the-art methods for identifying fake news in terms of accuracy by 9% on average.

Link prediction (LP) in social networks is to infer if a link is likely to be formed in the future. Social networks (SN) are ubiquitous and have different types such as human interaction and protein-protein networks. LP uses heuristic methods including common neighbors and resource allocation to find the formation of future links. These heuristics are sensitive to different types of social networks. Certain types of heuristics work better for some SN types, but not for others. Selecting the appropriate heuristic method for the SN type is often a trial and error process. Recent ground-breaking methods, WLMN and SEAL, demonstrated that this selection process can be automated for the different types of SN. While these methods are promising, in some types of SN they still suffer from low accuracy. The objective of this paper is to address this weakness by introducing a novel framework called PLACN that incorporates the analysis of common neighbors of nodes on target link and a combination of heuristic features through a deep learning method. PLACN is driven by a new method to efficiently extract the subgraphs for a target link based on the common neighbors. Another novelty is the method for labeling subgraphs based on the average hop and average weight. Furthermore, we introduce a method to evaluate the approximate number of nodes in the subgraph. Our model converts link prediction to an image classification problem and uses a convolutional neural network. We tested our model on seven real-world networks and compared against traditional LP methods as well as two recent state-of-the-art methods based on subgraphs. Our results outperformed those LP methods reaching above 96% of AUC in benchmark SNs.

Link prediction and node classification in social networks remain open research problems with respect to Artificial Intelligence (AI). Innate representations about social network structures can be effectively harnessed for training AI models in a bid to predict ties; and detect clusters via classification of actors with regard to a given social network. In this paper, we have proposed a distinct hybrid model: Representation Learning via Knowledge-Graph Embeddings and Convolution Operations (RLVECO), which hybridizes the strengths of Knowledge-Graph Embeddings (VE) and Convolution Operations (CO) in extracting and learning meaningful features from social graphs via Representation Learning (RL). RLVECO utilizes an edge sampling approach for exploiting features of a social graph via learning the context of each actor with respect to its neighboring actors.

Social Network Analysis (SNA) has become a very interesting research topic with regard to Artificial Intelligence (AI) because a wide range of activities, comprising animate and inanimate entities, can be examined by means of social graphs. Consequently, classification and prediction tasks in SNA remain open problems with respect to AI. Latent representations about social graphs can be effectively exploited for training AI models in a bid to detect clusters via classification of actors as well as predict ties with regard to a given social network. The inherent representations of a social graph are relevant to understanding the nature and dynamics of a given social network. Thus, our research work proposes a unique hybrid model: Representation Learning via Knowledge-Graph Embeddings and ConvNet (RLVECN). RLVECN is designed for studying and extracting meaningful representations from social graphs to aid in node classification, community detection, and link prediction problems. RLVECN utilizes an edge sampling approach for exploiting features of the social graph via learning the context of each actor with respect to its neighboring actors.

Predicting the future link between nodes is a significant problem in social network analysis, known as Link Prediction (LP). Recently, dynamic network link prediction has attracted many researchers due to its valuable real-world applications. However, most methods fail to perform satisfying prediction accuracy in various types of networks because the dynamic LP in evolving networks is struggling with spatial and nonlinear transitional patterns. Besides this, existing methods mostly involve the whole network and target link for the LP process. It leads to high computational costs. This paper aims to address these issues by proposing a novel framework named DLP-LES using deep learning methods. DLP-LES uses common neighbors based subgraph of a target link and learns the transitional pattern of it for a given dynamic network. We extract a set of heuristic features of the evolving subgraph to gather additional information about the target link. In this way, we avoid examining the entire network. Additionally, our model introduces new mechanisms to reduce computational costs. DLP- LES generates a lookup table to keep the required information of links of the network and uses a hashing method to store and fetch link information. We propose an algorithm to construct feature matrices of the evolving subgraph to learn transitional link patterns. Our model transforms the dynamic link prediction to a video classification problem, and uses Convolutional Neural Networks with Long Short-Term Memory neural networks. To verify the effectiveness of DLP-LES, extensive experiments are carried out on five real-world dynamic networks. We compare those results against four network embedding methods and basic heuristic methods.

Infectious diseases can spread from an infected person to a susceptible person through direct or indirect physical contact, consequently controlling such types of spread is difficult. However, a proper decision at the initial stage can help control the disease's propagation before it turns into a pandemic. Social distancing and hospital capacity are considered among the most critical parameters to manage these types of conditions. In this paper, we used artificial agent-based simulation modeling to identify the importance of social distancing and hospitals' capacity in terms of the number of beds to shorten the length of an outbreak and reduce the total number of infections and deaths during an epidemic. After simulating the model based on different scenarios in a small artificial society, we learned that shorter social isolation activation delay has a higher impact on reducing the catastrophe. Increasing the hospital's treatment capacity, i.e., the number of isolation beds in the hospitals can become handy when social isolation cannot be activated shortly. The model can be considered a prototype to take proper steps based on the simulations on different parameter settings towards the control of an epidemic.

In recent times, Social Network Analysis (SNA) has become a very important and interesting subject matter with regard to Artificial Intelligence (AI) in that a vast variety of processes, comprising animate and inanimate entities, can be examined by means of SNA. As a result, prediction tasks within social network structures have become significant research problems in SNA. Hidden facts and details about social network structures can be effectively and efficiently harnessed for training AI models with the goal of predicting missing links/ties within a given social network. Thus, important factors such as the individual attributes of spatial social actors, and the underlying patterns of relationship binding these social actors must be taken into consideration because these factors are relevant in understanding the nature and dynamics of a given social network structure. In this paper, we have proposed an interesting hybrid model: Representation Learning via Knowledge-Graph Embeddings and ConvNet (RLVECN). Our proposition herein is designed for examining, extracting, and learning meaningful facts for resolving link prediction problems about social network structures. RLVECN utilizes an edge sampling approach for exploiting the representations of a social graph, via learning the context of each actor with respect to its neighboring actors, with the goal of generating vector-space embeddings per actor which are further harnessed for innate representations via a Convolutional Neural Network (ConvNet) sublayer. Successively, these relatively low-dimensional representations are fed as input features to a downstream classifier for solving link prediction problems in a given social network. Our proposition, RLVECN, has been trained and evaluated on six (6) real-world benchmark social graph datasets.

Node classification and link prediction problems in Social Network Analysis (SNA) remain open research problems with respect to Artificial Intelligence (AI). Inherent representations about social network structures can be effectively harnessed for training AI models in a bid to detect clusters via classification of actors as well as predict ties with regard to a given social network. In this paper, we have proposed a unique hybrid model: Representation Learning via Knowledge-Graph Embeddings and ConvNet (RLVECN). Our proposition is designed for analyzing and extracting expressive feature representations from social network structures to aid in link prediction, node classification and community detection tasks. RLVECN utilizes an edge sampling technique for exploiting features of a given social network via learning the context of each actor with respect to its associate actors.

Event prediction in social network structures is a crucial research problem in social network analysis. This impels understanding the intrinsic relationship patterns preserving a given social network structure, based on the study of several structural properties computed on the constituent social units, with respect to space and time. In this regard, tackling problems of this nature is considered NP-Complete. Consequently, this paper proposes an original and unique approach which involves making event predictions about a target social unit, y, based on the intrinsic patterns of relationship learnt from one or more neighboring social units. Our methodology is based on Deep Learning (DL) architectures, and is developed using deep-layer stacks of Multilayer Perceptron (MLP) appended with an adjustment-bias (ab) vector at the output layer in a bid to improve the accuracy and precision of predictions made with respect to the target unit (or node). Also, we trained and tested our technique on a real world social clique comprising 5 connected cities; thereafter, we performed a comparative analysis of our approach against 9 other models drawn from the fields of Deep Learning, Machine Learning, and Statistics.

Mining medical data images have great potential for exploring hidden patterns in the medical domain. Medical data are heterogeneous which involves images to a great extent like MRI, ECG or Stroke effects etc. Knowledge discovery from such data can improve the diagnostic technique. However, to make the machine learn from such datasets requires large data. In the low-data regime, machine learning algorithms work poorly. Data Augmentation alleviates this by using existing data more effectively, but standard data augmentation produces only limited alternative data. Recent developments in Deep Learning field is noteworthy when it comes to learning probability distribution of points through neural networks, and one of key part for such progress is because of Generative Adversarial Networks(GANs). In this paper, we propose an evolutionary training technique using a cultural algorithm(CA) for neuro-evolution of deep task oriented GANs to find the best architecture for the given dataset. This architecture will help in generating similar but completely new data images which can be further used for training diagnostic Neural Networks. We have compared our approach with the Genetic Algorithm(GA) based neuro-evolution of GANs and show that CA based neuro-evolution of GANs evolves architecture which can generate a higher number of stroke-face images with better resolution when there is low data of original stroke faces.

In this paper, a novel knowledge-based evolutionary algorithm is proposed to assemble a team of care providers for patients in community-oriented palliative care. The main objective of this research is to optimize the patient's care services and human resource allocation process. From a system perspective in palliative care, there exists a group of patients with needs who are not able to perform some of their ordinary life activities due to their limited capability, as a consequence of their disease or disorders. On the other hand, we have a group of care providers who are capable, skilled, and ready to provide a wide range of services to the patients to fulfill those needs. This poses the challenge of assigning members to a team of care providers in an optimal manner to help the patient satisfy their needs, while taking into consideration the communication, distance and contact costs. To deal with this problem, we propose a novel algorithm based on a cultural algorithm (CA) as the basis for our model for assembling an optimal team of care providers. The overall goals are to minimize the costs and increase the patient's satisfaction rate. We have evaluated our model using multiple synthetic networks and conducted comparative analysis with other existing methods. The results show that our proposed model can overcome the shortcomings posed by the existing approaches.

A new architecture for Multi-Population Cultural Algorithm is proposed which incorporates a new Multilevel Selection framework (ML-MPCA). The approach used in this paper is based on biological group selection theory which aims to improve the capability of MPCA to tackle evolution of cooperation. A two-level selection process is introduced namely within-group selection and between-group selection. Individuals interact with the other members of the group in an evolutionary game that determines their fitness. If the group reaches a certain size, it splits into two daughter groups. We test our algorithm on CEC 2015 expensive benchmark functions to evaluate its performance. We show that our proposed algorithm improves solution accuracy and consistency. The model can be extended to more than two levels of selection and can also include migration.

A social health care system, and palliative care in particular, can be viewed as a social network of interacting patients and care providers. Each patient in the network has a set of capabilities to perform his or her intended daily tasks. However, some patients may not have the required capabilities to carry out their desired tasks. Consequently, different groups of care providers offer the patients support by providing them with a variety of needed services. Assuming there is a cost and resource limitations for providing care within the system, where each care provider can support a limited number of patients, the problem is to find a set of suitable care providers to match the needs of the maximum number of patients. In this paper, we propose a novel agent-based model to address this problem by extending the agent's capabilities using the benefit of the social network. Our assumption is that each agent, or patient, can cover its disabilities and perform its desired tasks through collaboration with other agents, or care providers, in the network. The goal of this work is to improve the quality of services in the network at both individual and system levels. On one hand, an individual patient wants to maximize his/her goals, while at the system level we want to achieve quality care for as many patients as possible with minimum cost. The performance and functionality of this proposed model have been evaluated based on various synthetic networks. The results demonstrate a significant reduction in the operational costs and enhancement of the service quality.

Multi-Population Cultural Algorithms (MPCA) define a set of individuals, each belonging to one of a set of populations that determines the shared goal, behavior or knowledge space of the individual. The design of MPCA extends from Cultural Algorithms (CA), which in turn improves on Genetic Algorithms (GA). To date, all examples of MPCA restrict individuals to belong to a single population at any given time, which can limit search potential and ability to simulate scenarios inspired by observations in real life. This article adopts ancestral “Heritage” as a new paradigm to extend MPCA. We introduce the “Heritage-Dynamic Cultural Algorithm” (HDCA) to allow easier definition of heterogeneous individuals and encourage greater search potential. As a test case, HDCA is compared directly with versions of MPCA, CA and GA against single-objective numerical optimization functions to demonstrate these intentions and inspire its use for new applications.

Networks can be analyzed from different aspects-micro and macro. If we assume that the main asset of each network is its population and the key difference between populations is their knowledge, then it is knowledge that drives the evolution of any network. In this paper, the behavior and status of a network will be analyzed in a case where a population from one network migrates to another similar network and transfers its knowledge to it. In fact, we are going to find how a migrated population will adapt itself to a new environment with similar characteristics based on the knowledge that it has learned from the previous network and what is the role of this prior knowledge in its evolution. For this purpose, different scenarios are modeled by employing a cultural algorithm with various networks and populations on two different cases: a population with migrated knowledge and a population without it. The results clearly show that when the changes in the structure of networks are less than 25%, trained population can adapt itself with the new network very fast but when the difference is higher, in the best case they perform like a random population without any training.

In this article, we introduce a computational model for the appraisal processes underlying emotion regulation from the perspective of Affect Control Theory. According to this theory, the affective meaning of emotions, behaviours, objects, and other entities can be assessed and projected onto a three dimensional space of evaluation, potency, and activity. This concept was applied to events occurring in the environment of an affective agent in order to study the dynamics of emotional changes caused by these events. Several appraisal processes were used to effectively analyze the affective impact of the occurred events and to interpret them in terms of the three dimensions of the affect control theory. A fuzzy automata framework was investigated and found to be a good fit to represent the dynamics of changes in the affective states and to effectively picture the transitions between different emotional response levels. Based on the results obtained from conducted experiments, we can argue that the proposed model has the potential to be used in predicting the emotional changes in (human/virtual) agents as a result of the occurrence of emotion-triggering events. Furthermore, it would appear that the proposed model has the capability to be used as the kernel of an extended system developed for reverse engineering events, and to generate and apply those in-favor of emotion regulation process.

In this study we introduce a new method to enable the migration of individuals from one population to another using the concept of dominance in Multi-Population Cultural Algorithms (MPCA's). The MPCA's artificial population comprises of agents that belong to a certain sub-population. Multiple sub-populations are generated, each running its own Cultural Algorithm (CA). In this work we create a dominance-MPCA (D-MPCA) with a network of populations that implements a dominance strategy. We hypothesize that the evolutionary advantage of dominance can help improve the performance of MPCA in general optimization problems. The Sphere function from the CEC 2013 benchmark optimization functions is used to calculate the fitness value of the individuals. We observe how the populations adapt to the changes. Preliminary results show improved performance in our proposed D-MPCA over traditional MPCA.

Multi-population cultural algorithms are cultural evolutionary frameworks involving multiple independently evolving subpopulations. Artifact selection involves the ability of agents to autonomously reason about selecting artifacts towards achieving their goals. In this study, agent migration between populations in a multi-population cultural algorithm is explored as an approach for augmenting artifact selection knowledge in social agents. Embedded in a social simulation model the multipopulation cultural algorithm consists of two subpopulations where agents in one subpopulation consistently outperform agents in the other due to the presence of knowledge about certain artifacts. Social networks connect agents within a subpopulation and agent knowledge can be altered by members of their network or the best performers of their subpopulation. The model investigates agent migration with novel artifact knowledge from the advanced subpopulation to the underperforming one. Child safety restraint selection is provided as an implemented case study. Results demonstrate the benefits of migration with a higher likelihood of an increase in agent performance when the social network is enabled. The study shows that culturally evolving agents can improve artifact selection knowledge in the absence of standard interventions as a result of migration.

Dynamic Heterogeneous Multi-Population Cultural Algorithm (D-HMP-CA) is a novel optimization algorithm which presents an effective as well as efficient performance to solve large scale global optimization problems. It incorporates dynamic decomposition techniques in order to divide problem dimensions among its local CAs. The variable interactions is not considered in the incorporated dynamic decomposition techniques. In this article, a new strategy is incorporated to detect the variable interactions to improve the process of dimension decomposition. This strategy is integrated into bottom-up dynamic decomposition technique and the integration is called supervised bottom-up approach. The proposed approach is evaluated over the large scale global optimization problems. The evaluation results reveal that the proposed approach outperforms the classical bottom-up technique in solving separable and single-group non-separable optimization functions, while the classical bottom-up approach offers a better performance for multi-group non-separable functions. However, the proposed supervised bottom-up approach presents a more efficient performance compared to the classical bottom-up method which shows that the variable interaction detection strategy does not impose extra computational costs.

This paper proposes using a fuzzy state machine to model the transition process among different levels, ranging from extreme to neutral, for any given emotion. The fuzzy transition function of this automaton is based on a three dimensional analysis of the events that take place in the system. The Pleasure, Arousal and Dominance (PAD) evaluation vector encodes rich information about the levels of pleasure, arousal and dominance respectively that are associated with the occurred event. The proposed model demonstrates the construction of the full automaton for transitions over different emotional states. It further tracks the current emotional state of an individual after injecting a series of events to the system. The model ultimately identifies the most reliable transition sequence between a pair of given initial and final target states.

Considering the dynamic and complex nature of real systems, it is not easy to build an accurate artificial simulation. Agent Based Modeling Simulations used to build such simulated models are often oversimplified and not realistic enough to predict reliable results. In addition to this, the validation of such Agent Based Model (ABM) involves great difficulties thus putting a question mark on their effective usage and acceptability. One of the major problems affecting the reliability of ABM addressed in this work is the dynamic nature of the environment. An ABM initially validated at a given time stamp is bound to become invalid with the inevitable change in the environment over time. Thus, an ABM that does not learn regularly from its environment cannot sustain its validity over a longer period of time. It should therefore have the ability to absorb changes in the environment upon their detection. Thus, in this paper we present a novel approach for incorporating adaptability and learning in an ABM simulation, thereby making it capable to be consistently synchronized with the changing environment and provide reliable results. One phase of our method explores the use of Data Mining (DM) in ABM for detecting environment trends and dynamics. Another phase addresses different methods for finding similarity between the knowledge represented by two different decision trees, for detecting a change in the simulation's environment.

In this article, a new architecture for Cultural Algorithms is proposed. The new architecture incorporates a number of sub-populations such that each sub-population is designed to optimize different parameters. According to the assigned parameters, each sub-population is a set of partial solutions which are managed by a local CA. Local CAs do not communicate with each other directly. In this architecture, a shared belief space is considered to record the best parameters. Local CAs send their best partial solutions to the belief space every generation. The belief space then updates its record of best parameters which will be used later by local CAs to evaluate their partial solutions. Due to incorporating a number of heterogeneous sub-populations, the proposed architecture is called Heterogeneous Multi-Population Cultural Algorithm (HMP-CA). Additionally, a local search heuristic is proposed to speed up the convergence of HMP-CA. The proposed HMP-CA is evaluated using a number of numerical optimization benchmark functions. The results show that the HMP-CA without the local search offers competitive results compared to the state-of-the-art methods and incorporating the proposed local search heuristic makes the proposed HMP-CA more efficient such that it outperforms all the state-of-the-art methods.

The subject of artifact or tool use is considered in many fields to be a vital area of research in the study of general human competence. Recently in artificial intelligence, formalizations of the mental attitudes of intentional agents have been extended to include agent capabilities with respect to artifacts or tools. We consider understanding how these individual capabilities are learned and how they evolve as important steps towards formally defining, representing and implementing complex group capabilities. In this paper, a theoretical model for artifact capability is extended to incorporate evolution and learning through exploratory methods. A representation of artifacts and the cognition of a rational agent that can learn artifact use are provided. Supervised learning is assumed and combined with historical knowledge and genetic algorithms to provide an implementation of a multi-agent simulation. The simulation is built to support an agent with the ability to learn an artifact capability through observations of its own behavior, as well as through observations of other agents in a social environment. Results obtained from the simple yet practical approach, show that learned use of artifacts outperforms random use and rational agents can learn artifact use more efficiently as a social species than on their own.

In this article, a new Memetic Algorithm (MA) has been proposed to solve Job Shop Scheduling Problems. The proposed MA is based on Machine Operation Lists (MOL), which is the exact sequence of operations for each machine. Machine Operation Lists representation is a modification of Preference List-Based representation. Linear Order Crossover (LOX) and Random operations are first considered as crossover and mutation operators for the proposed MA. Local Search heuristic (LS) of the proposed MA reconsiders all the operations of a job. It chooses a job and removes all of its operations and finally reassigns them again one by one in their sequencing order to improve the fitness value of the schedule. The proposed algorithm has been applied on the well-known benchmark of classical Job Shop Scheduling Problems (JSSP). Comparing it with the existing methods shows that the proposed MA and the proposed Genetic Algorithm (GA) without LS are effective in JSSP. Moreover, comparing the results of MA and GA shows that using LS not only improves the final results but also helps GA to converge to the final solution.

Social agent simulations are typically highly dynamic and complex multi-agent based models. Different methodologies exist to enable computer models to accomplish agent collaboration. Reputation models emerge as promising methods to control the communication framework between social agents. In this study we evaluate the social behaviour of agents guided by simulated awareness of self and others, where ethical constraints, stress avoidance and emotional interference play a role in the decision making process. A multi-agent based graph colouring model is formulated and extended to enable the social paradigm. Experiments using standard sets present an application neutral platform in order to study the effects of conscience decision making of the social agents on problem convergence when solving the graph colouring problem.

Specialization, or division of labour, leads to increased productivity in systems. We study the effect of social influence on the level of agent specialization in complex systems connected via social networks. There are several methods that explain the emergence of specialization, with the most prominent being the genetic threshold model. This model posits that agents possess an inherent threshold for task stimulus, and when that threshold is exceeded, the agent will perform that task. The idea of social influence is that an agent's choice of which task to specialize in when multiple ones are availabe, is influenced by the choices of its neighbours. Using the threshold model and an established metric that quantifies the level of agent specialization, we found that social influence leads to an increase in the division of labour.

Game playing has always provided an exciting avenue of research in Artificial Intelligence. Various methodologies and techniques have been developed to build intelligent game players. Coevolution has proven to be successful in learning how to play games with no prior game knowledge. In this paper we develop a coevolutionary system for the General Game Playing framework, where absolutely nothing is known about the game beforehand, and enhance it using Cultural Algorithms. In order to test the effectiveness of complementing coevolution with cultural algorithms, we play matches in several games between our player, a random player and one trained using standard coevolution.

In previous work we introduced an agent based model prototype to simulate the effect of the use of vehicle restraint systems on injury levels in children passengers. The goal is to better understand driver behavior in deciding on using, and correctly selecting, the constraint type and matching location in the vehicle for the child. In this study we enable drivers to learn from their individual driving experience with an option for immediate feedback from an expert intervention source following an accident. After implementing a social network that enables a driver to identify kin and neighbor relations in its community, the driver then may reciprocate this knowledge either by affecting related individuals or being influenced by others' beliefs. A Cultural Algorithm is designed to enable population level learning and to capture dominant social beliefs among drivers. Situational knowledge is implemented in the belief space based on top performing exemplars; those with demonstrated skillful use of child restraints following an accident with the least injury outcome are retained. We show that the presence of a cultural belief system positively influences the system performance measured in terms of the correctness of the use of child vehicle safety restraint and the injury level in children from vehicle accidents. However, the presence of culture made the population more resilient to changes after an intervention. This suggests that culture plays an important role in carrying out a successful intervention by health care practitioners.

With the ubiquity of e-commerce, process oriented websites have increased in complexity associated with the growing volume of transactions and underlying business rules. A single business process involves a careful execution of rules in a specified order. Current conceptual design for such websites consists of a data, hypertext and presentation layer. It is in the hypertext layer where both navigation and functional elements are embedded. The functional elements are in turn rich in business rules that are hidden in the navigation elements. Consequently, there are no clear rule definitions, and thereby leading to erroneous results in websites generated from these models.In this study, we isolate the business rules from the hypertext layer and introduce a new process definition layer to capture these rules. The proposed layer acts as a centralized repository of well-defined rules and applies them against corresponding executable processes. We propose a hybrid model framework that combines WebML, a web modeling language for representing the informational elements, and W3C's Web Services Choreography Description Language (WS-CDL), a platform independent portable language based on XML for representing functional elements that encapsulates the business rules. We construct a sample bargain store case study rich in business rules in order to test the proposed model. In the implementation phase we use the Pi4SOA tool for the realization of the now isolated business rules written in WS-CDL. For each executed business process initiated by an associated scenario file, the process definition layer verifies it against the corresponding rules and returns an appropriate response accordingly. This approach reveals that WS-CDL can successfully be used to implement business rules. Moreover, the proposed hybrid approach enables the separation of business rules from the hypertext layer, and thereby provides an enhanced conceptual framework to better handle process oriented applications.

In recent work we introduced an agent based model prototype to enable intervention studies by simulating the effect of the use of vehicle restraint systems on injury levels in children passengers. In a socially motivated dynamic framework, modeled drivers, or agents, are able to identify kin and neighbor relations. A knowledge structure to capture the driver's knowledge of the perceived correct child seat selection and location configuration is implemented. In this study, using a cultural algorithm, with situational knowledge dominant in the belief space, we enable both positive and negative exemplars in order to guide the belief at the population level. Based on evolving individual experiences, and corresponding changes in the belief system, the presence of both positive and negative exemplars were shown to be influential on the overall children population health and improved driver correctness in selecting the correct child restraint.

The village multi-agent simulation relieves the years of early Pueblo Indian settlers from A.D. 600 to 1300. The objective is to investigate why these settlers abandoned the region. Initial work modeled environmental aspects, such as water and paleoproductivity data, and farming practices of the households or agents. A cultural algorithm is then implemented to model the social networks and learning of exchange practices that evolved across these networks. Kinship, economic and community networks emerged. In this study, we introduce protein resources from simulated deer, hares and rabbits. Next, we enable the agents to hunt for these resources in order to survive. Furthermore, agents who cannot acquire sufficient resources from hunting may invoke their social networks and learn to exchange needed resources. As a result, protein resources presented a stress on the population that may motivate them to exit the region in search of better hunting grounds.

In this paper, we use principles from game theory, computer gaming, and evolutionary computation to produce a framework for investigating one of the great mysteries of the ancient Americas: why did the pre-Hispanic Pueblo (Anasazi) peoples leave large portions of their territories in the late A.D. 1200s? The gaming concept is overlaid on a large-scale agent-based simulation of the Anasazi. Agents in this game use a cultural algorithm framework to modify their finite-state automata (FSA) controllers following the work of Fogel (1966). In the game, there can be two kinds of active agents: scripted and unscripted. Unscripted agents attempt to maximize their survivability, whereas scripted agents can be used to test the impact that various pure and compound strategies for cooperation and defection have on the social structures produced by the overall system. The goal of our experiments here is to determine the extent to which cooperation and competition need to be present among the agent households in order to produce a population structure and spatial distribution similar to what has been observed archaeologically. We do this by embedding a "trust in networks" game within the simulation. In this game, agents can choose from three pure strategies: defect, trust, and inspect. This game does not have a pure Nash equilibrium but instead has a mixed strategy Nash equilibrium such that a certain proportion of the population uses each at every time step, where the proportion relates to the quality of the signal used by the inspectors to predict defection. We use the cultural algorithm to help us determine what the mix of strategies might have been like in the prehistoric population. The simulation results indeed suggest a mixed strategy consisting of defectors, inspectors, and trustors was necessary to produce results compatible with the archaeological data. It is suggested that the presence of defectors derives from the unreliability of the signal which increases under drought conditions and produced increased stress on Anasazi communities and may have contributed to their departure.

The multi-agent village simulation was initially developed to examine the settlement and farming practices of prehispanic Pueblo Indians of the Central Mesa Verde region of Southwest Colorado (Kohler, 2000; Kohler et al.). The original model of Kohler was used to examine whether drought alone was responsible for the departure of the prehispanic Puebloan people from the Four Corners region after 700 years of occupation. The results suggested that other factors besides precipitation were important. We then proceeded to add economic factors into the simulation, first allowing agents to engage in reciprocal exchanges between kin. This resulted in larger populations, more complex social networks, and more resilient systems. However, the exchange was done randomly and individuals did not remember the transactions. In This work we explicitly embed the reciprocal exchange process within a cultural algorithm, where individual agents can remember individuals that they have cooperated with. Also, in the cultural space the group can learn generalizations about what kind of relative is likely to successfully respond to a request. These generalizations are used to drive changes in requestor behavior. The results of this approach produced an even larger and more complex system exhibiting greater dependence on hub nodes that are sensitive to precipitation.

Explanations for the collapse of complex social systems including social, political, and economic factors have been suggested. Here we add cultural factors into an agent-based model developed by Kohler for the Mesa Verde Prehispanic Pueblo region. We employ a framework for modeling cultural evolution, cultural algorithms developed by Reynolds (1979). Our approach investigates the impact that the emergent properties of a complex system will have on its resiliency as well as on its potential for collapse. That is, if the system's social structure is brittle, any factor that is able to exploit this fragility can cause a collapse of the system. In particular, we will investigate the impact that environmental variability in the Mesa Verde had on the formation of social networks among agents. Specifically we look at how the spatial distribution of rainfall impacts the systems structure. We show that the distribution of agricultural resources is conducive to the generation of so called "small world" networks that require "conduits" or some agents of larger interconnectivity to link the small worlds together. Experiments show that there is a major decrease in these conduits in early 1200 A.D. This can have s serious potential impact on the networks resiliency. While the simulation shows an upturn near the start of the 14th century it is possible that the damage to the network had already been done.