Server deployment is a fundamental task in mobile edge computing: where to place the edge servers and what user cells to assign to them. To make this decision is context-specific, but common goals are (1) computing efficiency: maximize the amount of workload processed by the edge, and (2) communication efficiency: minimize the communication cost between the cells and their assigned servers. We focus on practical scenarios where the user workload in each cell is unknown and time-varying, and so are the effective capacities of the servers. Our research problem is to choose a subset of candidate servers and assign them to the user cells such that the above goals are sustainably achieved under the above uncertainties. We formulate this problem as a stochastic bilevel optimization, which is strongly NP-hard and unseen in the literature. By approximating the objective function with submodular functions, we can utilize state-of-the-art greedy algorithms for submodular maximization to effectively solve our problem. We evaluate the proposed algorithm using real-world data, showing its superiority to alternative methods; the improvement can be as high as 55%.

Memristor-based in-memory computing has emerged as a promising paradigm to overcome the constraints of the von Neumann bottleneck and the memory wall by enabling fully parallelisable and energy-efficient vector-matrix multiplications. We investigate the effect of nonlinear, memristor-driven weight updates on the convergence behaviour of neural networks trained with equilibrium propagation (EqProp). Six memristor models were characterised by their voltage-current hysteresis and integrated into the EBANA framework for evaluation on two benchmark classification tasks. EqProp can achieve robust convergence under nonlinear weight updates, provided that memristors exhibit a sufficiently wide resistance range of at least an order of magnitude.

With the rapid development of wireless communication has made efficient spectrum assignment a crucial factor in enhancing network performance. As a combinatorial optimization model for channel assignment, the radio labeling is recognized as an NP-hard problem. Therefore, converting the spectrum assignment problem into the radio labeling of graphs and studying the radio labeling of specific graph classes is of great significance. For G, a radio labeling φ : V ( G ) → { 0 , 1 , 2 , … } is required to satisfy | φ ( u ) − φ ( v ) | ≥ diam ( G ) + 1 − d G ( u , v ) , where diam ( G ) and d G ( u , v ) are diameter and distance between u and v. For a radio labeling φ, its span is defined as the largest integer assigned by φ to the vertices of G; the radio labeling specifically denotes the labeling with the minimal span among possible radio labeling. The strong product is a crucial tool for constructing regular networks, and studying its radio labeling is necessary for the design of optimal channel assignment in wireless networks. Within this manuscript, we discuss the radio labeling of strong prismatic network with star, present the relevant theorems and examples, and propose a parallel algorithm to improve computational efficiency in large-scale network scenarios.

This research work introduces a hybrid Network-on-Chip (NoC) topology routing algorithm that combines Odd-Even (OE) deterministic routing and Traffic-Aware Adaptive Routing (TAAR) to optimize data transmission efficiency in neuromorphic processors. The new routing scheme successfully reduces the latency and network congestion using both distance-based and dynamic congestion-aware path selection algorithms. Using the dependability of OE and the flexibility of TAAR, the system can transmit packets more balanced and more efficiently over the NoC. The algorithm itself uses Manhattan distance as the basis for computation and adds a congestion-aware cost function to dynamically route around hotspots in the network. The whole architecture is deployed on the Zynq UltraScale+ ZCU106 FPGA with better computational efficiency and lower power use. Synthesis and simulation in Verilog by Vivado 2024.1 ensure the correctness and real-time compatibility of the design. The hybrid routing method proposed herein provides a drastic improvement of communication performance, making the proposed method extremely suitable for AI-based and neuromorphic computing applications.

The spanning edge-cyclability is a generalization of hamiltonicity and pancyclicity of graphs. For a set F of r independent edges of G, if there exist r vertex-disjoint cycles that together span V ( G ) and each cycle contains exactly one edge of F, then G is called a spanning r-edge cyclable graph. In this paper, we study the spanning edge-cyclability of the n-dimensional augmented cube A Q n , which is a hypercube generalization. We prove that A Q n is spanning r-edge cyclable for 1 ≤ r ≤ n − 3 , if the set F of r independent edges of A Q n contains at least one complement edge.

Discrete time stochastic and deterministic Petri box calculus (dtsdPBC) is a parallel process algebra with stochastic and deterministic delays. To evaluate performance in dtsdPBC, semi-Markov chains (SMCs) and (reduced) discrete time Markov chains (DTMCs/RDTMCs) are analyzed. Stochastic bisimulation equivalence is used for quotienting the transition systems, SMCs and DTMCs/RDTMCs of the process expressions while preserving stationary behaviour and residence time. Our example of generalized shared memory system with maintenance demonstrates modeling, performance analysis and reduction by quotienting. The generalized system takes the probabilities and weights from the standard system's specification as variables, adjusted for performance optimization.

The ability to detect defective insulators has recently made remarkable progress. However, the accuracy and reliability of the traditional detection methods remain largely unexplored, especially in low-quality signal-modal data settings. In this paper, a novel image-text multimodal fusion deep learning network for detecting insulator defects is proposed to improve insulator defect detection accuracy. Multimodal fusion integrates text and image data information to enhance the understanding and analysis of insulator defects. This method introduces the transformer framework with six group attention mechanisms to integrate the image and text features. To extract fine image features, a feature pyramid network for multi-scale feature extraction is employed to extract defect features of varying sizes. Then, we incorporate feature differencing and principal component analysis-based colour feature enhancement to better capture fine details and contextual information features. We employ the WordPiece method to process the text data to obtain the text feature information. The experimental results indicate that the proposed method surpasses traditional approaches, significantly enhancing both accuracy and robustness in complex environments. Our model shows a substantial increase in recall rate: it is 14% higher than Simple Siamese, 18% higher than bootstrap your own latent (BYOL), and 31.7% higher than contrastive learning of visual representations (SimCLR).

Quality of transmission of optical signals, which are propagated beyond their transparent reach, is ensured by re-amplification, re-timing and re-shaping (3R-regeneration) at certain node(s) designated as regeneration site(s). Inclusion of regeneration sites increases capital expenditure (CAPEX) of Translucent Elastic Optical Networks (T-EONs) and so Regeneration Sites Selection (RSS) is an important problem to address. Survivability of T-EONs is a vital issue as a fiber-cut (link failure) can cause disruption of huge traffic. In this paper, we address the RSS problem for survivable T-EONs (ST-EONs). We formulate RSS as an offline problem for ST-EON design. We propose an Integer Linear Program (ILP) formulation for small networks and also heuristic algorithms for large networks. Results from simulation experiments performed on three network topologies suggest that the proposed algorithms can minimize CAPEX for ST-EON design.

This paper deals with the development of a new cellular automata-based model to describe the dynamics of direct selling, multilevel marketing companies. While many cellular automata models have been developed in the field of theoretical marketing, direct selling with multilevel marketing has not been addressed up to now. This specific marketing technique requires to account for both spatial spreading and the internal dynamics of each cell, in which six types of individuals are defined. The associated populations in each cell evolve using a new proposed aggregated-level model made of discrete time equations based on a mix of classical social contagion models and Boltzmann-inspired relaxation-reaction models. The proposed model is able to account for spatial spreading via weak social ties on a Moore neighbourhood, along with internal dynamics (growth, decay and equilibrium) of the six interacting populations in each cell. The model incorporates user-defined parameters that allow to model the dynamics of a given company via parameter fitting.

The study of complex networks has garnered significant attention in recent years due to their ability to visualize intricate systems. However, understanding the fundamental structures underlying their formation dynamics has remained elusive due to the enigmatic interplay between local and global properties. A hallmark of real-world complex networks is the presence of power-law distributions, which account for many of their intriguing characteristics. Preferential attachment has emerged as a common mechanism for generating scale-free networks. This research introduces a novel strategy for creating scale-free networks by capturing the spatiotemporal properties of an Elementary Cellular Automata trajectory. The process has been validated through the analysis of information-theoretic measures such as correlation information and entropy rate. Our findings demonstrate how the system’s sub-linear movement towards entropy reduction and increased correlations leads to state space localization, manifesting as a power-law distribution in the resulting network. The proposed method implicitly offers various forms of preferential attachment (sub-linear, linear, and super-linear) based on the sampling time pattern, circumventing the need for the first-mover advantage to generate hubs and requiring no global system information. We evaluate the resulting network using well-established criteria.