Representative feature selection is pivotal for cancer risk prognosis, as it directly determines the accuracy and clinical interpretability of diagnostic models. However, it remains challenging due to medical datasets' inherent high dimensionality, small sample size, and class imbalance. Existing methods focus primarily on statistical correlations between features and cancer risk, ignoring the critical trade-off between correlation and intrinsic causality-impairing clinical interpretability essential for guiding clinical decisions. To address this, we propose a constraint-based causal feature selection strategy using the causal Markov condition to discover internal causal relationships between cancer risk types and features, overcoming limitations of traditional associative methods. Specifically, a causal model is built on a network skeleton constructed by integrating improved neighborhood component analysis (NCA) and conditional independence tests (G2 test) to filter redundancy and retain causally relevant nodes. Theoretical analysis confirms the model adheres to the causal Markov property, preserving local causal relationships and reflecting global feature-cancer risk dependencies. Additionally, a graph-constrained estimator quantifies each feature's causal effect on risk labels to identify the most representative and discriminative features for prognosis. Empirical validation on two in-house clinical datasets with cancer images and two public datasets (a breast cancer dataset with image-derived cellular features and a high-dimensional speech signal-based clinical dataset) demonstrates our approach outperforms competing methods including traditional machine learning and mainstream causal feature selection algorithms in prediction accuracy, sensitivity and specificity, confirming its effectiveness and reliability for clinical cancer risk prognosis.

Effective agent coordination is crucial in cooperative Multiagent Reinforcement Learning (MARL). While recent advances have significantly improved cooperation by modeling agent interactions through various graph structures, most existing approaches primarily focus on homogeneous agents. Despite the ubiquity of heterogeneous agents, constructing a comprehensive graph that captures their diverse attributes and relationships from scratch is notoriously labor-intensive for both humans and agents, which makes policy learning extremely challenging. To tackle this difficulty, we propose a novel method that utilizes a fuzzy human attention-guided graph to model inter-agent relationships. Instead of learning the graph entirely from scratch, we incorporate abstract human attention, with its uncertainty captured through fuzzy logic, to guide the graph development process. To further accommodate the varying attributes and objectives of heterogeneous agents while maintaining their learning capabilities, the attention-guided graph is fine-tuned through a hyper-network. Our proposed approach is end-to-end trainable and agnostic to specific MARL methods. Empirical evaluations conducted on challenging heterogeneous scenarios from the StarCraft Multiagent Challenge (SMAC) and SMACv2 validate the effectiveness of the proposed method.

Identifying representative causal features from computed tomography (CT) images remains a significant challenge for the subtype classification of non-small cell lung cancer (NSCLC). Existing methods, whether based on radiomics or deep neural networks, often overlook the intricate causal relationships among features, thereby yielding suboptimal or even detrimental diagnostic outcomes. To bridge this gap, we propose a Hybrid Representative Causal Network (HRCL) for NSCLC subtype identification, which explicitly captures the local causal relationships inherent in the interaction between radiomics and features based on deep learning from a holistic perspective. Specifically, a causal network structure is learned to delineate the unique causal configuration of distinct NSCLC subtypes through a variable number of nodes and links. The resultant network adheres to the causal Markov property, thereby ensuring global consistency of all local cause-effect dependencies. Moreover, a hybrid representative feature selector is designed to identify the most salient causal features from the causal network for precise NSCLC subtype classification. Our method achieves an accuracy of 83.7% on the publicly available P-NSCLC dataset and 90.3% on the privately collected I-NSCLC dataset. The empirical evaluations demonstrate that our model significantly outperforms the state-of-the-art methods.

Cognitive load significantly affects performance in safety-sensitive fields like aviation, where excessive cognitive demands are linked to increased operational errors and heightened risks. However, cognitive load measurement faces the challenge of neglecting spatio-temporal and causal dependencies in multi-channel physiological signals. We present a score-based causal representation model that captures causal structural diversity and spatiotemporal dependencies within single channels and inter-dependencies among multiple channels. By optimizing a score-based causal function under causal Markov property constraints, the model disentangles latent spatiotemporal features into causal and task-irrelevant groups. Experimental results on two public and one in-house dataset show our model significantly outperforms state-of-the-art methods.

Deep reinforcement learning (DRL) has emerged as a powerful paradigm for solving complex decision-making problems. However, DRL-based systems still face significant dependability challenges particularly in real-time environments due to the simulation-to-reality gap, out-of-distribution observations, and the critical impact of latency. Latency-induced faults, in particular, can lead to unsafe or unstable behaviour, yet existing fault-tolerance approaches to DRL systems lack formal methods to rigorously analyse and optimise performance and safety simultaneously in real-time settings. To address this, we propose a formal framework for designing and analysing real-time switching mechanisms between DRL agents and alternative controllers. Our approach leverages Timed Automata (TAs) for explicit switch logic design, which is then syntactically converted to a Markov Decision Process (MDP) for formal analysis. We develop a novel convex query technique for multi-objective model checking, enabling the optimisation of soft performance objectives while ensuring hard safety constraints for MDPs. Furthermore, we present MOPMC, a GPU-accelerated software tool implementing this technique, demonstrating superior scalability in both model size and objective numbers.

Due to its exceptional learning ability, multi-agent deep reinforcement learning (MADRL) has garnered widespread research interest. However, since the learning is data-driven and involves sampling from millions of steps, training a large number of agents is inherently challenging and inefficient. Inspired by the human learning process, we aim to transfer knowledge from humans to avoid starting from scratch. Given the growing emphasis on the Human-on-the-Loop concept, this study focuses on addressing the challenges of large-population learning by incorporating suboptimal human knowledge into the cooperative multi-agent environment. To leverage human experience, we integrate human knowledge into the training process of MADRL, representing it in natural language rather than specific action-state pairs. Compared to previous works, we further consider the attributes of transferred knowledge to assess its impact on algorithm scalability. Additionally, we examine several features of knowledge mapping to effectively convert human knowledge to the action space where agent learning occurs. In reaction to the disparity in knowledge construction between humans and agents, our approach allows agents to decide freely which portions of the state space to leverage human knowledge. From the challenging domains of the StarCraft Multi-agent Challenge, our method successfully alleviates the scalability issue in MADRL. Furthermore, we find that, despite individual-type knowledge significantly accelerating the training process, cooperative-type knowledge is more desirable for addressing a large agent population. We hope this study provides valuable insights into applying and mapping human knowledge, ultimately enhancing the interpretability of agent behavior.

Cognitive load recognition faces the challenge of capturing the complex physiological events and their causal relationships in real time. Existing psychological experiment-based approaches, such as those using NASA-TLX and SWAT scales, provide subjective assessments but fail to objectively track changes in dynamic cognitive load. In contrast, while data-driven deep learning models automate feature extraction from physiological signals, they often overlook the causal relationships between features, leading to less interpretable results. To address these limitations, we introduce a multi-channel spatio-temporal causal network model that integrates Granger causality (GC) and Gramian angular difference fields (GADF) learned from multiple physiological data for cognitive load recognition. Our model utilizes a GC network to delineate temporal causalities within cognitive states and GADF maps to capture spatio-temporal dependencies among physiological signals. These features are fused using a capsule network, preserving local causal and spatio-temporal integrity. Comprehensive evaluations on both publicly available benchmarks and proprietary VR datasets demonstrate that the proposed framework achieves state-of-the-art performance in cognitive load monitoring. The system exhibits exceptional capability in identifying sophisticated physiological signatures and their inherent causal dependencies.

The imperative for coordination among intelligent machines has popularized cooperative multiagent reinforcement learning (MARL) in AI research. However, compared to the well-explored homogeneous agent cooperation, heterogeneous agents with different attributes or behaviors are more prevalent in practical scenarios yet they have received relatively little attention. Due to the heterogeneity of agents and the diversity of relationships, learning efficient coordination among heterogeneous agents is particularly challenging, suffering from the curse of dimensionality and the start-up problem. To tackle these challenges, we propose a novel method that connects humans and agents under a hierarchical structure to guide the learning of MARL agents. Drawing inspiration from knowledge transfer among diverse human individuals, we consider human attention as a general pattern that can be applied to heterogeneous agents. Instead of relying on comprehensive step-by-step demonstrations, we utilize fuzzy logic to capture the abstraction and vagueness within suboptimal human guidance. To avoid negative knowledge transfer, we leverage human attention as an auxiliary source through hyper-networks, allowing agents to selectively adapt to the proposed human prior knowledge. The proposed method is agnostic to specific MARL methods and can be flexibly integrated with diverse algorithms. We conduct experiments on challenging tasks within the StarCraft Multiagent Challenge (SMAC) and SMACv2 environments, and the empirical results demonstrate that our method can improve existing methods in several heterogeneous scenarios.

Stream processing frameworks have become key enablers of real-time data processing in modern distributed systems. However, robust and scalable mechanisms for verifying temporal properties are often lacking in existing systems. To address this gap, a new runtime verification framework is proposed that integrates linear temporal logic (LTL) monitoring into stream processing applications, such as Apache Spark. The approach introduces reusable LTL monitoring patterns designed for seamless integration into existing streaming workflows. Our case study, applied to real-time financial data monitoring, demonstrates that LTL-based monitoring can effectively detect violations of safety and liveness properties while maintaining stable latency. A performance evaluation reveals that although the approach introduces computational overhead, it scales effectively with increasing data volume. The proposed framework extends beyond financial data processing and is applicable to domains such as real-time equipment failure detection, financial fraud monitoring, and industrial IoT analytics. These findings demonstrate the feasibility of real-time LTL monitoring in large-scale stream processing environments while highlighting trade-offs between verification accuracy, scalability, and system overhead.

INTRODUCTION:Patients with alcohol use disorders have a high hospital re-admission rate, adding to the strain on the healthcare system. To address this issue, this study aimed to predict 28-day unplanned hospital re-admission for these patients. METHODS:From linked de-identified datasets, patients with alcohol use disorders who had hospital re-admissions between 2015 and 2018 were identified. Univariate and multiple logistic regression were conducted to select variables for inclusion in five machine learning models-logistic regression (baseline), random forest, support vector machine, long-short term memory and clinical bio bidirectional encoder representation of transformers (Clinical Bio-BERT)-to predict the 28-day re-admission. RESULTS:Eight hundred and sixty-nine patients with alcohol use disorders incurred 2254 hospital admissions. Patients aged 45-49 or 70-74 or 75-79 were 4-5 times more likely to be re-admitted than those in other age groups; males were 36% more likely than females; patients who use polysubstance were 3.3 times more likely than otherwise. Patients with "respiratory system disorders" or "hepatobiliary system and pancreas disorders" had 60% higher risk than otherwise. Interaction with emergency department or drug and alcohol service after discharge reduced the risk by 71% and 79%, respectively. The 10-variable Clinical Bio-BERT demonstrated the highest sensitivity (.724). DISCUSSION AND CONCLUSIONS:Patients with alcohol use disorders with the following characteristics were more likely to have unplanned re-admissions within 28 days: male, aged 45-49 or 70-74 or 75-79, with "respiratory system disorders" or "hepatobiliary system and pancreas disorders", or patients who use polysubstance. Interactions with emergency department or drug and alcohol service after discharge had reduced risk of hospital re-admission.

Multiagent reinforcement learning (MARL) has garnered extensive research attention due to its strong learning capabilities, leading to its deployment in increasingly challenging scenarios. Although progress has been made toward more generalizable solutions, many MARL algorithms continue to struggle with balancing scalability and heterogeneity, particularly under conditions of growing uncertainty. Research has shown that combining dense local interactions with sparse global interactions can significantly enhance scalability while preserving agent heterogeneity. Motivated by these insights and inspired by human social behavior, we propose a novel hierarchical method that integrates human guidance with multiagent systems (MASs). Rather than requiring agents to learn from scratch, our method transfers abstract knowledge from humans, employing fuzzy logic to manage the inherent uncertainty in this guidance and reduce the required human effort. To accommodate both local and global interactions, we introduce two levels of human guidance: individual action guidance for agents and an attention graph to describe agent relationships. Our proposed approach is end-to-end and compatible with diverse MARL algorithms. We evaluate our approach in the starcraft multiagent challenge (SMAC) and SMACv2 environments. Empirical results demonstrate its effectiveness, even under low-performance fuzzy human guidance.

The study of business process analysis and optimization has attracted significant scholarly interest in the recent past, due to its integral role in boosting organizational performance. A specific area of focus within this broader research field is Process Mining (PM). Its purpose is to extract knowledge and insights from event logs maintained by information systems, thereby discovering process models and identify process-related issues. The goal of the current study is to examine how Quantitative Model Checking (QMC) approaches might be applied in the context of PM. Model checking is a well-known verification approach that provides thorough analysis and validation of a system’s properties in comparison to a predetermined model. The adoption of QMC is aimed at improving the accuracy, reliability, and comprehensiveness of PM models in stochastic environment. We propose a novel methodology in this research direction, which integrates QMC with PM by formally modelling discovered and replayed process models and applying QMC methods to verify PM models. The potential of QMC to overcome significant drawbacks of the existing methodologies is the main driver for its use in PM. By including probabilistic model verification, it is possible to take into account the uncertainties and stochastic behaviour that are frequently present in systems that are used in real world; while statistical model checking methods utilized where probabilistic methods fails/not suitable, such as, to handle complex models and/or models with large state-spaces.

Workflows orchestrate a collection of computing tasks to form a complex workflow logic. Different from the traditional monolithic workflow management systems, modern workflow systems often manifest high throughput, concurrency and scalability. As service-based systems, execution time monitoring is an important part of maintaining the performance for those systems. We developed a trace profiling approach that leverages quantitative verification (also known as probabilistic model checking) to analyse complex time metrics for workflow traces. The strength of probabilistic model checking lies in the ability of expressing various temporal properties for a stochastic system model and performing automated quantitative verification. We employ semi-Makrov chains (SMCs) as the formal model and consider the first passage times (FPT) measures in the SMCs. Our approach maintains simple mergeable data summaries of the workflow executions and computes the moment parameters for FPT efficiently. We describe an application of our approach to AWS Step Functions, a notable workflow web service. An empirical evaluation shows that our approach is efficient for computer high-order FPT moments for sizeable workflows in practice. It can compute up to the fourth moment for a large workflow model with 10,000 states within 70 s.

In multi-alternative elections, messages in support of or against alternatives spread across online social media. To make decisions, individuals integrate various sources of information, including information from their friends in online social networks. As information may continuously update online, opinions of individuals also become dynamic, making it challenging to predict voting outcomes. In this paper, we propose a novel voting metric named minimal influence gap, which is based on the information structure induced within a social network, to predict voting outcomes in multi-alternative elections. We evaluate this metric as a predictor of voting outcomes across three popular models of voting behaviours: the Maximising Expected Utility model, the Local Dominance model, and the K-Pragmatist model. We test our metric on synthetic networks with a scale-free feature or a community structure, and on a real-world social network. Experimental results demonstrate that the minimal influence gap strongly correlates with voting outcomes in the three models, particularly in the Local Dominance model when voters persist in supporting their favourite alternative.

Most research investigating how social influence affects election results mainly uses diffusion models for binary opinions. However, these diffusion models are progressive and focus on the diffusion of one opinion. In this article, we introduce a general diffusion model for ordinal opinions expressed as linear orderings over a finite set of candidates. We employ agent-based modeling to simulate a nonprogressive diffusion process, allowing multiple types of opinion diffusion about different candidates. The proposed agent-based diffusion model can forecast long-term trends of opinion diffusion in social networks by capturing voters’ personalized features and incorporating dynamic social contexts. Furthermore, we examine the possibility of affecting election outcomes by externally changing the ordinal opinions of certain vertices, i.e., campaigning. Since finding influential voters from the social network is computationally challenging, we propose a heuristic approach, i.e., backward influence rank (BIR). Experimental results demonstrate that the proposed BIR approach is superior to the classic greedy approach for campaigning by achieving a similar margin of victory to that of the greedy approach but running two orders of magnitude faster than the greedy approach did.

A key challenge in sensor-based fall prediction is the fact that a fall event can often occur in various configurations of fall poses together with their own spatio-temporal dependencies. This leads us to define a spatio-temporal model to explicitly characterize these internal configurations of poses. In particular, we introduce a graph neural network with spatio-temporal topological structure to encode such latent relations among poses by capturing representative patterns in fall events. Moreover, a human body orientation estimator is devised to capture human low limbs information, and as a result, separate pose dependencies are globally consistent. Empirical evaluations on two benchmark datasets and one in-house dataset suggest our approach significantly outperforms the state-of-the-art methods.

Agents with limited capacities need to cooperate with others to fulfil complex tasks in a multi-agent system. To find a reliable partner, agents with insufficient experience have to seek advice from advisors. Currently, most models are rating-based, aggregating advisors' information on partners and calculating averaged results. These models have some drawbacks, like being vulnerable to unfair ratings under a high ratio of dishonest advisors or dynamic attacks and locally convergent. Therefore, this paper proposes a Ranking-based Partner Selection (RPS) model, which clusters honest and dishonest advisors into different groups based on their different rankings of trustees. Besides, RPS uses a sliding-window-based method to find dishonest advisors with dynamic attack behaviours. Furthermore, RPS utilizes an online-learning method to update model parameters based on real-time interaction results. According to experiment results, RPS outperforms ITEA under different kinds of unfair rating attacks, especially in two situations: 1) there is a high ratio of dishonest advisors; 2)dishonest advisor takes dynamic attack strategies.

VarietyDetect, a semi-supervised model combining transfer learning and self-training, enhances anomaly detection across domains. Leveraging both labeled and unlabeled data, VarietyDetect demonstrates a 24.6% F1 score improvement on hate speech detection and reduces false negatives by 15% on average. Its adaptability and iterative self-training position it as a robust solution for real-world short-text anomaly detection.

Download This Paper Open PDF in Browser Add Paper to My Library Share: Permalink Using these links will ensure access to this page indefinitely Copy URL Copy DOI