This paper investigates remote state estimation in cyber-physical systems (CPS) where direct and reliable communication between distributed sensors and the control center is unavailable due to the lack of infrastructure. In such scenarios, unmanned aerial vehicles (UAVs) can be deployed as mobile data collectors to facilitate information transfer. While joint UAV trajectory and resource allocation problems have been widely studied in various contexts, adapting these to minimize estimation error in remote state estimation of linear CPS—accounting for the nonlinear evolution of error covariances and spatio-temporal coupling—presents distinct challenges. To tackle this, we formulate a joint optimization problem and decompose it into UAV path planning and UAV-sensor resource allocation. The path planning problem is modeled as a Markov Decision Process, where we establish the existence and periodicity of the optimal policy, and the resource allocation is formulated as a many-to-one matching game. We jointly solve these sub-problems via an alternating optimization algorithm that combines dynamic programming with a deferred acceptance-based matching scheme, and we prove the stability of the resulting matching. Simulation results confirm the periodic structure of the optimal path and demonstrate the effectiveness of the proposed method in reducing estimation error compared with other representative matching-based schemes.

Federated learning (FL) enables collaborative model training across distributed edge devices without sharing raw data, but the large number of devices and frequent parameter exchanges create substantial communication overhead. To improve communication efficiency, event-triggered mechanisms have been introduced. However, such approaches may exclude valuable but inactive users (i.e., IoT devices with informative data but limited activity) or overlook subtle but important updates, resulting in suboptimal trade-offs between communication efficiency and training performance. To address this, we propose Stochastic Event-Triggered Lazy Aggregation with Stochastic Quantization (SET-LASQ). SET-LASQ employs a probabilistic triggering strategy that adaptively balances the participation of highly contributing users with the reactivation of long-inactive ones, while incorporating gradient quantization to further reduce transmitted bits. Under strong convexity, we theoretically prove that SET-LASQ achieves sublinear convergence with a diminishing stepsize and linear convergence to a steady-state neighborhood with a fixed stepsize. Extensive experiments on benchmark datasets verify that SET-LASQ reduces both communication rounds and transmission volume while maintaining competitive model accuracy, consistently outperforming deterministic event-triggered and lazy aggregation baselines under independent and non-independent data distributions.

3D point cloud semantic segmentation (PCSS) is a cornerstone for environmental perception in robotic systems and autonomous driving, enabling precise scene understanding through point-wise classification. While unsupervised domain adaptation (UDA) mitigates label scarcity in PCSS, existing methods critically overlook the inherent vulnerability to real-world perturbations (e.g., snow, fog, rain) and adversarial distortions. This work first identifies two intrinsic limitations that undermine current PCSS-UDA robustness: (a) unsupervised features overlap from unaligned boundaries in shared-class regions and (b) feature structure erosion caused by domain-invariant learning that suppresses target-specific patterns. To address the proposed problems, we propose a tripartite framework consisting of: 1) a robustness evaluation model quantifying resilience against adversarial attack/corruption types through robustness metrics; 2) an invertible attention alignment module (IAAM) enabling bidirectional domain mapping while preserving discriminative structure via attention-guided overlap suppression; and 3) a quality-guided contrastive memory bank that progressively refines pseudo-labels with feature quality for more discriminative representations. Extensive experiments on SynLiDAR-to-SemanticPOSS adaptation demonstrate a maximum mIoU improvement of 14.3% under adversarial attack.

This paper studies the online reward poisoning problem, wherein an adversary deliberately manipulates the reward function during training to mislead the learning agent into adopting a mischievous policy. While the majority of existing reward poisoning research focuses on offline attacks, which assume prior knowledge of the transition probability, our work explores a more practical yet challenging dynamics-agnostic scenario. Specifically, we consider the scenario where the adversary has access to the agent’s replay buffer and can modify the reward data without the knowledge of transition probabilities. We formalize the poisoning task as an optimization problem and employ a reformulation method to circumvent the double-sampling issue. The proposed algorithm is provably convergent in the tabular setting and can be extended to the function approximation setting, where the poisoned reward network and the poisoned Q-value network are jointly learned to solve the problem. The algorithm’s effectiveness is validated through four distinct experimental evaluations.

We investigate the capacity of an adversary to learn the underlying interaction network through repeated best response actions in linear-quadratic games. The adversary strategically perturbs the decisions of a set of action-compromised players and observes the sequential decisions of a set of action-leaked players. The central question pertains to whether such an adversary can fully reconstruct or effectively estimate the underlying interaction structure among the players. To begin with, we establish a series of results that characterize the learnability of the interaction graph from the adversary’s perspective by drawing connections between this network learning problem in games and classical system identification theory. Subsequently, taking into account the inherent stability and sparsity constraints inherent in the network interaction structure, we propose a stable and sparse system identification framework for learning the interaction graph based on complete player action observations. Moreover, we present a stable and sparse subspace identification framework for learning the interaction graph when only partially observed player actions are available. Finally, we demonstrate the efficacy of the proposed learning frameworks through numerical examples.

We study the denial-of-service (DoS) attack power allocation optimization in a multiprocess cyber–physical system (CPS), where sensors observe different dynamic processes and send the local estimated states to a remote estimator through wireless channels, while a DoS attacker allocates its attack power on different channels as interference to reduce the wireless transmission rates, and thus degrading the estimation accuracy of the remote estimator. We consider two attack optimization problems. One is to maximize the average estimation error of different processes, and the other is to maximize the minimal one. We formulate these problems as Markov decision processes (MDPs). Unlike the majority of existing works where the attacker is assumed to have complete knowledge of the CPS, we consider an attacker with no prior knowledge of the wireless channel model and the sensor information. To address this uncertainty issue and the curse of dimensionality, we provide a learning-based attack power allocation algorithm stemming from the double deep Q-network (DDQN) method. First, with a defined partial order, the maximal elements of the action space are determined. By investigating the characteristic of the MDP, we prove that the optimal attack allocations of both problems belong to the set of these elements. This property reduces the entire action space to a smaller subset and speeds up the learning algorithm. In addition, to further improve the data efficiency and learning performance, we propose two enhanced attack power allocation algorithms which add two auxiliary tasks of MDP transition estimation inspired by model-based reinforcement learning, i.e., the next state prediction and the current action estimation. Experimental results demonstrate the versatility and efficiency of the proposed algorithms in different system settings compared with other algorithms, such as the conventional value iteration, double Q-learning, and deep Q-network.

The past few years have witnessed the explosive growth of Internet of Things (IoT) devices. The necessity of real-time edge intelligence for IoT applications demands that decision making must take place right here right now at the network edge, thus dictating that a high percentage of the IoT created data should be stored and analyzed locally. However, the computing resources are constrained and the amount of local data is often very limited at edge nodes. To tackle these challenges, we propose a distributionally robust optimization (DRO)-based edge intelligence framework, which is based on an innovative synergy of cloud knowledge transfer and local learning. More specifically, the knowledge transfer from the cloud learning is in the form of a reference distribution and its associated uncertainty set. Further, based on its local data, the edge device constructs an uncertainty set centered around its empirical distribution. The edge learning problem is then cast as a DRO problem subject to the above two distribution uncertainty sets. Building on this framework, we investigate two problem formulations for DRO-based edge intelligence, where the uncertainty sets are constructed using the Kullback-Leibler divergence and the Wasserstein distance, respectively. Numerical results demonstrate the effectiveness of the proposed DRO-based framework.

In this work, we consider the multi-party privacy conflict (MPC) in an online social network (OSN). As many data items uploaded to the OSN are “co-owned” by multiple users with different privacy concerns, some personal information of OSN users may be disclosed by others unintentionally. On the contrary with existing mainstream OSN platforms allowing only the very user uploading the data to set the privacy level, in this article we take a fine-grained approach to resolve MPC, in which all co-owners independently determine whether to share their personal content within the data on OSN. Interacted with its peers, the opinion of a co-owner, however, might be influenced by and consequently influence the decision of its peers. To this end, each co-owner, as an individual decision maker, strikes a tradeoff between its internal privacy preference and the external social influence from its neighbors in a OSN. Specifically, we formulate the interaction among co-owners as a multi-player non-cooperative game with a network structure representing their social relations. For the proposed network game, we establish the existence of multiple (pure-strategy) equilibria, and characterize them accordingly. The convergence of interaction is also investigated when synchronous and asynchronous best-response updates are used, respectively. We note that when the action set for the players is discrete, the game exhibits non-linear dynamics, making it challenging to analyze the convergence behavior. We prove that synchronous update may lead to either an equilibrium or a strategy cycle, and the asynchronous update always leads to an equilibrium. Building upon this analysis, we advocate a practical implementation of the proposed MPC management, which balances the automation of the management and intervention of users. Moreover, we take one step further to develop approaches aiming to reach a “stronger agreement” among the players for the sake of benefits of uploader and OSN provider. Numerical examples are also provided to corroborate the analytical results.