This paper presents two-phase ordering (2PO), a new asynchronous transaction coordination algorithm that does not rely on a central coordinator. Each client constructs a view of the global execution graph and deterministically schedules its own transactions globally. We formally prove the correctness of 2PO and that the resulting schedules are conflict-serializable (CSR). In our measurements, 2PO exhibits high commit rates (above 97 % under high contention) while scaling linearly with the number of nodes.
Federated Learning (FL) offers a promising avenue for privacy-preserving Human Activity Recognition (HAR) using distributed on-body sensor data. However, existing FL approaches often underperform in real-world deployments due to client heterogeneity, modality-specific sensor dynamics, and the non-IID nature of human behaviors across users. We present FedMeet, a personalized federated learning framework for human activity recognition (HAR) using multi-sensor on-body sensing data from smartwatches and earables. FedMeet addresses key challenges in distributed HAR, including non-iid client data, sensor heterogeneity, and high intra-class variance with low inter-class variance, common in real-world deployments. Unlike existing methods, FedMeet introduces a unified architecture that integrates sensor fusion, temporal modeling, personalization, and privacy preservation. These gains stem from four key components: (1) gated sensor fusion for adaptive modality relevance, (2) a BiLSTM backbone to capture temporal activity patterns, (3) client-level personalization to accommodate behavioral and sensor placement heterogeneity, and (4) a privacy-preserving masking mechanism that securely perturbs client weight updates before transmission, reducing the risk of information leakage. Experimental results on a real-world multi-sensor HAR dataset show that FedMeet achieves a test accuracy of 87.97% within 20 rounds, outperforming state-of-the-art baselines (FedPer, FedProx, ClusterFL) across precision, recall, and F1-score.FedMeet offers a promising solution for privacy-aware, real-time HAR in collaborative environments such as workplaces and smart meeting spaces.
Collective perception via Vehicle-to-Network communications and multi-access edge computing (MEC) infrastructure offers a promising solution to the sensing limitations of individual vehicles. However, vehicles from different manufacturers often subscribe to different mobile network operators (MNOs), creating isolated perception silos across separate MEC infrastructures. This paper quantitatively evaluates MEC interoperability's impact on collective perception services using realistic vehicle mobility simulations in urban environment of Luxembourg City. We analyzed road object coverage and temporal freshness across distinct automakers and MNOs. Results demonstrate that synchronizing Local Dynamic Maps (LDMs) across different MNOs significantly improves road object perception coverage throughout all deployment phases. Also, age-of-information analysis confirms the critical need for low-latency inter-MEC communications.
The DISPERSION problem requires a set of k <= n autonomous robots, initially positioned arbitrarily on the nodes of an anonymous, port-labeled, n-node graph, to relocate such that at most one robot occupies each node. This paper investigates the DISPERSION problem in a torus topology of size root n x root n. Each node in the torus has its incident edges labeled with unique port numbers from the set {0, 1, 2, 3}. We consider the synchronous setting, where a robot can observe the memory of a robot located at the same node. We consider two models of the torus: (i) a oriented torus, where port labeling ensures consistent relative orientation, i.e., if p is the incoming port at a node v, then (p + 1) mod 4 denotes the next port in the anticlockwise direction; and (ii) an unoriented torus, where such relative directionality is absent. We present a distributed algorithm that solves the DISPERSION problem in both models in O(root n) rounds, without requiring robots to know n or k in advance. The algorithm achieves this using only O(log n) bits of memory per robot, which is asymptotically optimal. Our results generalize and extend the work of Kshemkalyani et al. [13], who considered DISPERSION on grid graphs under stronger assumptions. In particular, we show how to adapt the solution to operate effectively even in the absence of local orientation, addressing additional symmetry and coordination challenges inherent to the torus topology.
