The increasing demand for intelligent, privacy-aware, and scalable solutions at the edge of the network is accelerating the convergence of Giant AI models and Internet of Things (IoT) infrastructures in 6G environments. In this paper, we propose a symbiotic threat detection framework that unifies federated learning (FL), graph neural networks (GNNs), and homomorphic encryption to enable decentralized anomaly detection across distributed 6G-enabled IoT ecosystems. Our approach addresses key challenges in current cloud-centric architectures, including data privacy, communication efficiency, and lack of interpretability in AI-driven threat detection. The proposed framework, SymFL-GNN, supports collaborative learning among IoT devices while retaining data locally, leveraging the Paillier Homomorphic Cryptosystem to ensure gradient-level encryption. To enhance interpretability, a dynamic sensor graph is constructed using self-learned embeddings and attention mechanisms, allowing the model to pinpoint anomalous behaviors and their sources. We evaluate our framework on two real-world industrial datasets (SWaT and WADI) representing cyber-physical water systems, achieving 96.3% and 96.0% accuracy, respectively, with significant improvements over existing baselines in both precision and F1 score. Our results show that SymFL-GNN effectively balances local autonomy with global intelligence, supporting the vision of symbiotic AI at the edge. The framework demonstrates how 6G-enabled IoT networks can jointly contribute to and benefit from Giant AI models, laying the foundation for secure, intelligent, and privacy-preserving distributed systems in critical infrastructure and consumer environments.
13 February 2024 in IEEE Transactions on Computational Social Systems
Hate speech in social media is a growing problem that reinforces racial discrimination and mistrust between people, leading to physical crimes, violence, and fragmentation in world communities. Although previous studies showed the potential of user profiling in hate speech detection in social media, there has not been a thorough analysis of users’ characteristics and dispositions to understand the development of hate attitudes among users. To bridge this gap, we investigate the role of a wide range of psycholinguistic and behavioral traits in characterizing and distinguishing users prone to post hate speech on social media. Considering anti-Asian hate during the COVID-19 pandemic as a case study, we curate a dataset of 5 417 041 tweets from 3001 Twitter users prone to publish hate content (aka hateful-to-be users) and a corresponding matched set of 3001 control users. Our findings reveal significant statistical differences in most dimensions of psycholinguistic attributes and online activities of hateful-to-be users compared to control users. We further develop a classifier and demonstrate that features derived from user timelines are strong indicators for automatically predicting the onset of hateful behavior.
Software Defined Networking (SDN) has brought significant advancements in network management and programmability. However, this evolution has also heightened vulnerability to Advanced Persistent Threats (APTs), sophisticated and stealthy cyberattacks that current detection methods often fail to counter, especially in the face of zero-day exploits. A prevalent issue is the inadequacy of existing strategies to detect novel threats while addressing data privacy concerns in collaborative learning scenarios. This paper presents P3GNN, a novel model that synergizes Federated Learning (FL) with Graph Convolutional Networks (GCN) for autonomous and effective APT detection in SDN environments. P3GNN autonomously analyzes operational patterns within provenance graphs, identifying deviations indicative of security breaches. Its core feature is the integration of FL with homomorphic encryption, which fortifies data confidentiality and gradient integrity during collaborative learning. This approach addresses the critical challenge of data privacy in shared learning contexts. Key innovations of P3GNN include its ability to autonomously detect anomalies at the node level within provenance graphs, offering a detailed view of attack trajectories. Furthermore, the model's unsupervised learning capability enables it to independently identify zero-day attacks by learning standard operational patterns. Empirical evaluations using the DARPA TCE3 dataset demonstrate P3GNN's exceptional performance, achieving an accuracy of 0.93 and a low false positive rate of 0.06.