Neuro-symbolic artificial intelligence (NSAI) is a novel paradigm that combines the pattern recognition capabilities of neural networks with the logical reasoning of symbolic systems [11, 13]. This study explores the application of NSAI for detecting fraudulent activities on the Ethereum blockchain, employing an existing implementation known as Logic Tensor Networks (LTNs) [12, 13]. The task is formulated as a binary classification aimed at distinguishing between legitimate and fraudulent transactions. Ethereum is a decentralized, open-source blockchain, peer-to-peer platform based on the Kademlia protocol and powered by the Ethereum Virtual Machine (EVM), which enables the creation and execution of smart contracts and decentralized applications (dApps). Smart contracts are self-executing programs on the blockchain that automatically verify and enforce the terms of an agreement in a trustless manner. Once deployed, they become immutable and publicly auditable on the blockchain. The decentralized and pseudonymous nature of Ethereum presents significant challenges for effective fraud detection. In this study, we propose and evaluate a voting-based ensemble NSAI framework to classify fraudulent transactions using a publicly available Ethereum fraud detection dataset [14]. Experimental results demonstrate that the proposed ensemble architecture outperforms approximately 80
Structural health monitoring (SHM) systems for large-scale infrastructures often rely on dense sensor networks, which are prone to faults, generate high-volume data, and require computationally efficient algorithms to ensure low-latency inference for real-time monitoring. To enhance overall network accuracy and robustness, aggregating the predictions of individual sensors provides a way to leverage complementary information across the network while mitigating sensor-level errors. In this study, we investigate intelligent aggregation strategies for single-sensor classifiers in SHM networks. We leverage acceleration time series data from the RT345 bridge dataset, collected from a real instrumented structure, to detect and classify structural damages. Individual sensor classifiers produce probabilistic predictions, which are then combined using different aggregation strategies. Soft averaging serves as a baseline, while stacking ensembles employs linear meta-classifiers (Logistic Regression) for interpretable per-sensor weighting and nonlinear meta-classifiers (Random Forest) to capture complex conditional dependencies across sensors, albeit at the cost of interpretability and stability. Experimental results demonstrate that meta-learning strategies significantly improve classification accuracy and robustness. We further evaluate prediction time, scalability, and model size, highlighting trade-offs between linear and nonlinear aggregation for real-time SHM applications. Finally, we extend the study by exploring alternative acceleration time series representations, showing that system efficiency can be improved without compromising damage detection performance.
The rapid proliferation of Internet of Things devices introduces significant challenges in determining optimal service placement across heterogeneous computing layers, spanning from centralized cloud servers to decentralized fog and edge nodes. Effective placement is essential to minimize latency, reduce energy consumption, and control costs, all while adhering to resource constraints such as limited memory and processing capacity. In this paper, we investigate several strategies for solving the IoT service placement problem. To overcome the limited adaptability of traditional optimization approaches, we introduce dynamic resource management in the simulation environment YAFS and propose a Deep Reinforcement Learning–based approach utilizing a Double Deep Q-Network (DDQN) architecture. The DRL agent autonomously learns placement policies through continuous interaction with the environment, optimizing a weighted multi-objective reward that balances execution time, energy efficiency, and cost. Experimental evaluations were conducted under two scenarios: memory-only constraints and combined memory plus CPU constraints. Results demonstrate that the DRL-based strategy consistently outperforms baseline approaches, including three meta-heuristic optimization methods—Genetic Algorithm, Simulated Annealing, and Tabu Search—as well as the cloud-only strategy, across all performance metrics.
The proliferation of vehicle-to-everything (V2X) communication systems with diverse quality of service (QoS) requirements has created an increasingly complex heterogeneous wireless environment where vehicles must intelligently select among multiple radio access technologies (RATs) to meet these diverse and stringent QoS requirements. In this paper, we address the critical challenge of optimal RAT selection in multi-RAT vehicle to infrastructure (V2I) communication. We propose an intelligent adaptive RAT selection algorithm that combines reinforcement learning with predictive mobility modeling to enable real-time decision making under uncertainty and incomplete information scenarios with a view to simultaneously optimizing latency and reliability while reducing the number of handovers under rapidly changing vehicular network conditions. To validate our approach, we conduct extensive network simulations and the results demonstrate significant improvements compared to state-of-the-art RAT selection method: up to 63.0
By converting high-level user objectives into workable configurations, intent-based networking, or IBN, seeks to automate network administration. Predicting future intentions to facilitate proactive resource management is a major difficulty in IBN. The use of Long Short-Term Memory (LSTM) networks for network intent prediction is examined in this research. Using the Network Intent Language (NILE), we create a synthetic dataset of 8,353 intents and assess how well the LSTM model performs across a range of temporal window sizes. Our findings show that an LSTM model with a medium window size (e.g., 12) may successfully strike a compromise between stability and responsiveness, which makes it appropriate for real-world intent forecasting tasks like resource allocation in 5G networks and proactive service orchestration.
