To advance the widespread application of quantum information technology, a network infrastructure is required to support the interconnection of numerous quantum nodes, enabling distributed quantum applications. Based on the fact that quantum nodes are interconnected by sharing entangled states, this paper proposes a tree-based networking scheme that takes entangled sources as the core of Quantum Local Area Networks (QLANs) to improve entanglement connectivity. Specifically, each entangled source serves as a branch point in the tree-based structure, efficiently enabling the physical and entanglementbased logical interconnection of multiple quantum nodes within a local area while reducing deployment costs. Furthermore, we introduce a dynamic entanglement distribution scheduling strategy and an entanglement connectivity-based path selection algorithm to promote the practical application of the proposed networking scheme, realizing efficient and robust entanglement distribution between distant quantum nodes in both inter-QLAN and intra-QLAN. Finally, we implement a quantum network with a backbone and ten tree-based QLANs on a simulation platform to evaluate our scheme. Extensive results demonstrate that the tree-based networking scheme can effectively promote the performance of long-distance entanglement distribution by interconnecting quantum nodes with high entanglement connectivity. Our work offers meaningful insights for building a global quantum Internet in the future.
We propose FE D-DPROC, a novel federated learning design principle to jointly provide differential privacy (DP), Byzantine robustness, and communication efficiency. Central to our approach is the concept of robust-compatible compression, which allows reducing the bi-directional communication overhead without undermining the robustness of the aggregation. We instantiate our design principle as RO BAJOL, which integrates the Johnson-Lindenstrauss (JL)-based compression mechanism with robust averaging for robustness. Our theoretical analysis establishes a convergence guarantee for FE D-DPROC under robust-compatible compression and additive Gaussian perturbations. We ensure that RO BAJOL maintains robustness guarantees, satisfies DP, and substantially reduces communication overhead. We further present simulation results on CIFAR-10, Fashion MNIST, and FEMNIST, evaluating the practical performance of RO BAJOL. We compare RO BAJOL with a state-of-the-art communication-efficient and robust FL scheme augmented with DP for a fair comparison, demonstrating that RO BAJOL outperforms existing methods in terms of robustness and utility under different Byzantine attacks.
We study asynchronous log-linear learning in weighted binary coordination games over binary symmetric and binary erasure channels. Snapshot communication applies the logit response to one noisy payoff, whereas fast communication applies it after channel-score averaging; the resulting kernels generally differ. With homogeneous links, the fast kernel is an exact Gibbs sampler for the original coordination potential at a reliability-scaled inverse temperature. Within the binary model, a heterogeneous or directional fast kernel is reversible if and only if its reliability-weighted couplings are reciprocal, yielding an edge-weighted effective potential. The snapshot kernel is ergodic but need not be reversible; as the inverse temperature β → 0, its local kernel ratios agree to first order with those of the fast sampler. A finite-K empirical-score model recovers snapshot updating at K = 1 and, for each fixed finite graph and β, converges pointwise in kernel and stationary law in the fast limit. Our simulation results illustrate the theoretical findings for a wide range of different channel parameters and network topologies.
Many emerging artificial intelligence (AI) applications continue to grow in model size and computational complexity. To meet these increasing requirements, distributed learning (DL) has become a key paradigm, enabling model training to be partitioned and executed across multiple geographically distributed datacenters (DC). However, DL inherently requires frequent synchronization of model parameters. As the number of participating local models and the size of AI models grow, DL generates substantial communication overhead, resulting in heavy demands on cross-DC optical networks. To solve this problem, as our main and original contribution, we propose a minimum spanning tree (MST)-based strategy that transforms the DL scheduling problem into a procedure based on the construction of multiple spanning trees. This approach enables flexible routing and model aggregation, effectively reducing communication costs in terms of both resource consumption and latency reduction. We further formulate an integer linear programming (ILP) model to obtain the optimal solutions for benchmarking. We evaluate the proposed methods on 9-node, 30-node and 68-node substrate networks, where the number of local models and DL requests are varied to provide a comprehensive assessment. Numerical results show that the MST-based strategy achieves near-optimal performance, with a gap of 1.1% to 3.99% compared to the optimal ILP solution. Meanwhile, the proposed MST strategy shows strong scalability and outperforms existing heuristic benchmarks in terms of resource efficiency and latency reduction.
