Ningjiang Chen

Dynamic directed acyclic graph (DAG) offloading in vehicular edge computing (VEC) must coordinate task precedence, heterogeneous queues, communication contention, and intermittent wireless contact. Because each placement changes the timing and availability of subsequent actions, fixed task orders and fixed-dimensional joint actors are poorly suited to this evolving decision process. HBF-GPPO is introduced to address this coupling through a joint policy that factorizes ready-task selection and task-conditioned resource placement. Directed graph message passing, an identifier-free set encoder, and action-conditioned task-resource features represent DAG structure, variable resource pools, and current timing feasibility. On nominal workloads, HBF-GPPO attains a mean penalized normalized makespan of 3.706 and a scheduling success rate of 97.12 %; its normalized-makespan reductions relative to PEFT and GA-DRL-A are 1.284 and 1.679, respectively. Across controlled changes in task count, resource count, and topology, it achieves a mean normalized makespan of 3.545 and a success rate of 95.24 %. Fixing the task order and removing action-conditioned state evolution increase normalized makespan by 0.833 and 3.641, respectively, showing the contribution of hierarchical decision making and state reconstruction.

CONTEXT:The rapid growth of logs in modern software systems makes log compression essential for reducing long-term storage costs. Existing methods mainly remove explicit structural redundancy via template extraction. As template redundancy becomes increasingly exploited, the compression bottleneck has shifted to log variables — dynamic data with heterogeneous types and time-varying drift. Current coding strategies struggle to model their complex conditional distributions, leaving residual redundancy and causing performance mismatch under traffic changes and system evolution.OBJECTIVES:This paper proposes LogLoom, a lossless log compression framework, to tackle the compression bottleneck caused by heterogeneous log variables and distribution drift in non-stationary scenarios.METHODS:LogLoom first decouples variable types to simplify probabilistic approximation across fields. It then combines type-aware reversible residual transformation with lightweight conditional probability coding, mapping high-entropy variables into lower-entropy representations without information loss. For online non-stationary settings, it further adopts drift-aware local adaptation to reduce model mismatch and stabilise compression performance.RESULTS:Experiments on the LogBase dataset show that LogLoom achieves an average compression ratio 2.8 times that of the best existing baseline, and delivers stronger robustness against temporal drift.CONCLUSION:Type-aware entropy reduction and local dynamic adaptation can effectively overcome current bottlenecks in lossless log compression, and are applicable to large-scale log storage scenarios with continuous system evolution.

In intelligent surveillance scenarios, the massive access of devices and complex task processing impose higher demands on real-time performance and resource scheduling in edge computing systems. To address the resulting challenges such as insufficient offloading robustness and low transmission efficiency, this paper proposes a two-stage intelligent offloading and hybrid transmission framework TSDM. The first stage employs hierarchical subspace-based reinforcement learning to enhance offloading robustness under dynamic and abrupt channel variations. The second stage introduces a Time-Division Asynchronous Hybrid Cooperative Multiple Access (TDA-HCMA) scheme, which dynamically partitions spectrum to balance spectral efficiency and low-delay transmission for critical tasks. To avoid local optima in multi-access scheduling, a bidirectional evolutionary hybrid genetic algorithm is designed for subchannel allocation optimization. By integrating reinforcement learning, adaptive multiple access, and intelligent scheduling, the proposed framework improves flexibility and robustness in multi-task edge offloading. Experimental results demonstrate superior performance in task completion, delay, and energy efficiency compared to existing methods in dynamic environments.

Purpose: In order to enhance the prediction accuracy of the grey model in small-sample, high-volatility scenarios and to effectively utilize spatial effects and data autocorrelation features, a weighted grey Markov forecasting model with spatial effects is constructed. Design/methodology/approach: A spatial effect module is incorporated to capture inter-regional correlations and heterogeneity, and a weighted Markov chain module is introduced, which adjusts the weights of different state transition steps to better utilize recent data. Findings: Experimental results on talent and electricity demand data from multiple regions show that the proposed model significantly outperforms baseline models in prediction accuracy, and ablation studies validate the effectiveness of both the spatial effect module and the weighted Markov chain module. Originality/value: Integrating spatial effects into the grey model and leveraging a weighted Markov chain to fully account for the autocorrelation characteristics of the data series, the proposed approach enhances both adaptability and predictive performance in highly volatile scenarios.

Traditional log parsing research has largely focused on single-line logs. However, real-world systems frequently generate hybrid logs composed of both single-line and multi-line entries. This paper addresses key challenges in hybrid log parsing, including the absence of standardized log definitions, benchmark datasets, and evaluation methodologies. To bridge these gaps, we introduce a formal annotation framework that leverages source code analysis and LLMs. This enables the construction of HybridLogHub, a standardized, large-scale hybrid log dataset that facilitates consistent and objective evaluation. In addition, we propose HybridTool, a context-aware parser that combines dynamic analysis with structural inference to effectively aggregate and parse multi-line logs. Experiments on 10 benchmark datasets demonstrate that our tool provides a reliable basis for evaluating existing parsers and highlights their strengths and limitations in hybrid scenarios. This work lays a foundation for future research by establishing unified evaluation standards and identifying critical challenges in hybrid log parsing.

In intelligent surveillance scenarios, the massive access of devices and complex task processing impose higher demands on real-time performance and resource scheduling in edge computing systems. To address the resulting challenges such as insufficient offloading robustness and low transmission efficiency, this paper proposes a two-stage intelligent offloading and hybrid transmission framework TSDM. The first stage employs hierarchical subspace-based reinforcement learning to enhance offloading robustness under dynamic and abrupt channel variations. The second stage introduces a Time-Division Asynchronous Hybrid Cooperative Multiple Access (TDA-HCMA) scheme, which dynamically partitions spectrum to balance spectral efficiency and low-delay transmission for critical tasks. To avoid local optima in multi-access scheduling, a bidirectional evolutionary hybrid genetic algorithm is designed for subchannel allocation optimization. By integrating reinforcement learning, adaptive multiple access, and intelligent scheduling, the proposed framework improves flexibility and robustness in multi-task edge offloading. Experimental results demonstrate superior performance in task completion, delay, and energy efficiency compared to existing methods in dynamic environments.

As a promising distributed computing paradigm, edge computing (EC) enhances service quality and reduces service latency by deploying computational and storage resources at the network edge. However, due to edge servers’ geographic distribution and resource constraints, EC is challenged by edge denial-of-service (EDDoS) attacks. Although various approaches have been proposed to mitigate EDDoS attacks, the impact of capacity uncertainties in edge server processing capacity and transmission latency has been largely overlooked. This limits the adaptability and robustness of mitigation strategies in edge computing environments, which could result in overload and paralysis of edge servers. To address this issue, this paper introduces uEDDoS-D, an uncertainty-aware EDDoS mitigation approach based on an improved deep deterministic policy gradient algorithm. uEDDoS-D models the uncertainties of computing capacities and transmission latency of edge servers in EC environments and leverages the collective computational resources of edge servers to mitigate EDDoS attacks without relying on traditional attack detection mechanisms. uEDDoS-D aims to minimize the impact of uncertainties in edge server processing capacity and transmission latency on mitigation decisions while effectively reducing service latency. Experimental results demonstrate that uEDDoS-D significantly reduces service latency, outperforming the state-of-the-art approaches by an average of 16.75%, and enhances system robustness in EDDoS attack scenarios.