In recent years, multiunmanned aerial vehicular systems (MUAVs) have become prevalent in divergent applications: agriculture, spectrum utilization, transportation, forest fire monitoring, and among others, due to their flexible, robust, and autonomous operational maneuver. Battery-powered multiunmanned aerial vehicles (MUAVs) systems possess limited computation and communication resources, significantly reducing their functional dimension by limiting mission time and range. To address this issue, we propose a federated deep reinforcement learning (FDRL)-based intelligent and decentralized task offloading scheme for resource-constrained UAVs that can enhance the operational capability of the MUAV systems. Moreover, the proposed FDRL scheme can improve offloading policy quality while preserving data privacy in MUAV. However, such intelligent systems may fall prey to backdoor attacks that can intervene in the system’s regular operation causing rapid degradation of its performance. We introduce a novel triggerless backdoor attack scheme on intelligent task offloading UAVs and analyze its impact to gauge the resiliency of the offloading policy in the presence of an adversary. Then, we propose lightweight agnostic defense mechanisms to combat such backdoors in multi-UAV settings. The extensive simulation results show that the proposed attack and defense strategies are practical and efficient.

The electric vehicle (EV) charging ecosystem, being a distinguishable paradigm of IIoT infrastructure, consists of distributed and complex hybrid systems that demand adaptive data-driven cyber-defense mechanisms to tackle the ever-growing attack vectors of cyber-physical systems. We propose an adaptive differential privacy-based federated learning framework for building a collaborative network intrusion detection system model for EV charging stations (EVCS). We use utility optimized local differential privacy to provide data privacy to the local network traffic data of each EVCS. Moreover, we propose a reinforcement learning-based intelligent privacy allocation mechanism at the EVCS level. The main significance of the proposed mechanism is that it can make privacy provisioning adaptive to the extent of privacy breaching rate, and dynamically optimize the privacy budget and the utility to avoid human intervention such as domain knowledge experts. The experimental results confirm the efficacy of our proposed mechanism and achieves appropriate privacy provisioning accuracy to approximately 95%.

Finding the root causes of network performance anomalies is critical to satisfy the quality of service requirements. In this paper, we introduce machine learning (ML) models to process TCP socket statistics to pinpoint underlying reasons of performance issues such as packet loss and jitter. More importantly, we introduce a novel feature engineering method to transform network-dependent metrics (e.g., total packet count and round trip time) in training datasets into network-independent forms to be able to transfer the models to new network settings without requiring to retrain them. Experimental results in various network settings show that the proposed feature engineering approach improves the performance of the models in previously unseen network settings from around 60% to nearly 90%. We believe ability to transfer ML models across networks will pave the way for wide adoption of ML solutions in production networks where collecting labeled data is not possible.

Federated learning (FL) has become an emerging machine learning technique lately due to its efficacy in safeguarding the client’s confidential information. Nevertheless, despite the inherent and additional privacy-preserving mechanisms (e.g., differential privacy, secure multi-party computation, etc.), the FL models are still vulnerable to various privacy-violating and security-compromising attacks (e.g., data or model poisoning) due to their numerous attack vectors which in turn, make the models either ineffective or suboptimal. Existing adversarial models focusing on untargeted model poisoning attacks are not enough stealthy and persistent at the same time because of their conflicting nature (large scale attacks are easier to detect and vice versa) and thus, remain an unsolved research problem in this adversarial learning paradigm. Considering this, in this paper, we analyze this adversarial learning process in an FL setting and show that a stealthy and persistent model poisoning attack can be conducted exploiting the differential noise. More specifically, we develop an unprecedented DP-exploited stealthy model poisoning (DeSMP) attack for FL models. Our empirical analysis on both the classification and regression tasks using two popular datasets reflects the effectiveness of the proposed DeSMP attack. Moreover, we develop a novel reinforcement learning (RL)-based defense strategy against such model poisoning attacks which can intelligently and dynamically select the privacy level of the FL models to minimize the DeSMP attack surface and facilitate the attack detection.

Vehicular edge computing (VEC), being a novel computing paradigm, promises to provide divergent vehicular edge services, both functional (e.g., charging route prediction, emergency messages, etc.) and infotainment (e.g. video gaming applications, featured movie series, etc.), at the network edge while satisfying application-specific QoS requirements. Vehicles usually send these service requests to nearest roadside units (RSUs), which contain mobile edge servers, according to the functional requirements or the vehicle owner preferences. However, the VEC server's virtual resources may fall short compared to the unbounded amount of real-time service requests (infotainment/functional) during rush hours. This limitation entails VEC servers to fail to meet the stringent latency requirements which may create unwanted malfunction event during driving in the requested vehicles (if functional/critical service requests are delayed in processing). Moreover, the VEC environment's intrinsic properties, i.e. mobility, application-specific distinct latency requirements, traffic congestion, and uncertain task arrival rate, make the VEC task scheduling problem a non-trivial one. In this paper, we propose an extreme reinforcement learning (ERL) based context-aware VEC task scheduler that can make online adaptive scheduling decisions to meet the application-specific latency requirements for both types of tasks (i.e. functional and infotainment). The scheduler can make scheduling decisions directly from its experience without prior knowledge or the VEC environment model. Finally, we present extensive simulation results to confirm the efficacy of the proposed scheduler. Results show that the VEC server can achieve successful (by meeting QoS requirements) task completion rate of above 96% for different task arrival rates (ranging from 10 to 50 arrival/s) using the proposed scheduler. In the simulation, we also analyze the scheduling algorithm's scalability in response to the vertical expansion of the VEC server. Furthermore, we compare the performance of our proposed method with two baseline methods.