In microservice management, engineers analyze numerous heterogeneous performance metrics, including latency, CPU, and memory usage, to resolve incidents swiftly. Consequently, a metric-based root cause analysis (RCA) must demonstrate interpretability, accuracy, and efficiency. However, recent causal-discovery-based RCAs have been identified as having challenges in terms of accuracy and efficiency. This paper highlights the issues arising from the heterogeneity of metrics and attempts to improve causal-aware RCA while leveraging the metadata associated with the metrics. We adopt the validated concept from existing RCA methods that metrics showing greater deviations during a failure are likely the root causes. The proposed method quantifies metric deviations comparable among heterogeneous metrics and searches for propagation trees that best explain the observed deviations. Experimental results on microservice-based system benchmark datasets demonstrate that the proposed method outperforms existing RCA methods in terms of accuracy and efficiency.
With the widespread adoption of microservice architectures, system complexity has increased significantly, making fault root cause localization a critical issue in system operations. Under resource constraints or load pressure, the monitoring metrics that reflect system status often exhibit incompleteness, negatively impacting operational accuracy and system stability. To address this challenge, we propose ReconRCA, which consists of an offline reconstruction stage and an online localization stage. During reconstruction, ReconRCA leverages historical data, along with spatio-temporal models and attention mechanisms, to reconstruct missing metrics. In the online stage, it captures both intra and inter-metric correlations of microservices to achieve precise root cause localization. Experiments show that ReconRCA outperforms existing methods with both incomplete and complete data, achieving average top-1 hit rates at 33.28% and 39.74%, respectively.
Smart factories and homes represent resourceasymmetric environments where gateways offer abundant resources, while user and edge devices remain constrained. Existing authentication and key agreement (AKA) protocols for the settings typically overlook threats from malicious gateways and cannot reconcile dynamic anonymity with minimal overhead. To address these challenges, we propose the Secure dynamic Anonymity updates and Gateway-side secure Authentication (SAGA), a lightweight AKA protocol based on three synergistic layers: dual-threshold Shamir's Secret Sharing (SSS) to distribute trust across multiple gateways, a Merkle Tree mechanism with one-time password algorithm based on hash chains (HOTP) for efficient gateway verification, and an active shared key-refresh method to maintain dynamic anonymity and session keys. Formal security proofs using Burrows-Abadi-Needham (BAN) logic and ProVerif validate SAGA's rigorous security, while experimental results show it achieves stronger security properties with lower computational and communication costs than existing schemes.
Object-centric Predictive Monitoring has recently gained attention due to advances in machine learning and rise of Object-Centric Event Logs (OCELs), which comprehensively capture object interactions. This paper presents a modular framework supporting customizable pipelines for predictive analysis across diverse event logs. The framework comprises three core components: Preprocessing (preserving object relationships via graph structures), Graph Embedding Model, and Prediction Model. We experimentally evaluated various combinations of embeddings and predictors on three public OCELs. Results show that no single configuration consistently dominates. However, GAT and Graph Transformer models perform best for predicting remaining time and the number of events. Performance improves with larger embedding and subgraphs, particularly for neuralbased models. Finally, GAT delivered the most stable and highperforming results across all event logs in generalization tests.
Knowledge Graphs (KGs) have emerged as a critical technique to enhance recommendation performance by modeling complex relationships and semantics within heterogeneous networks. However, it faces issues such as longtail distribution, structural redundancy caused by semantically similar relations, and susceptibility to noise interference, which severely limit the effectiveness of graph-based recommendations. Aiming to tackle the challenges, we propose Relation-aware Contrastive Learning (RACL), a brand-new framework for knowledge-enhanced recommendations. Specifically, relationdriven subgraph construction is employed to cluster the KG into subgraphs with potential semantic associations, addressing the issue of structural redundancy while alleviating the long-tail effect through the integration of relationship types. Besides, we introduce a relation-aware aggregation module to inject relation-specific semantic features from KG into neighborhood propagation, effectively encoding multi-type relational contexts into user and item embeddings. Furthermore, a graph learner is established, which significantly improves the model's robustness in contexts with sparse and noisy data by integrating selfsupervised signals into model training. Comprehensive experiments on two publicly accessible datasets verify that our RACL surpasses the state-of-the-arts in terms of recommendation efficacy.
Chatbots have advanced from basic conversational agents to versatile tools by integrating external services. However, traditional chatbots are constrained by predefined service boundaries, limiting their ability to handle complex tasks with unintegrated services. While most research focuses on improving service discovery and invocation through data-intensive pretraining, only 13.29% of services are well-documented, hindering practical deployment. This paper proposes a self-improving workflow for chatbots, using a “wide in, strict out” self-supervised learning approach to acquire domain knowledge efficiently and generate high-quality service documents. Compatible with existing methods, it eliminates the need for dataset collection or pre-training. Experiments demonstrate that our workflow significantly improves the pass and success rate of chatbots in utilizing unintegrated services, offering a powerful solution for real-world applications where service integration is limited.
Deploying convolutional neural networks (CNNs) on resource-constrained Internet of Things (IoT) devices facilitates convenient intelligent services, which has attracted extensive attention. However, the dynamic resource availability poses significant challenges for resource-intensive CNNs. Recent advancements in efficient CNNs have two limitations: i) elevated resource consumption due to extensive datasets and post-training calibration; ii) lacking flexibility facing dynamic resources due to static model structure. To this end, we propose a RESourCe-Aware oncE-for-alL (ReScale) framework for efficient CNN deployment on IoT devices. Specifically, we propose a hierarchical filter generation mechanism to generate different amounts of filters dynamically, mapping a few learnable filters to abundant filters for discriminative feature extraction. Besides, we set a coefficient. to modulate the filter generation according to the available resources of IoT devices. With this framework, we only need to train CNN models once to handle different resource availability of IoT devices. Experimental results show that our proposed ReScale framework can generate more efficient models with lower resource consumption while maintaining high accuracy. Through the coefficient., our method enables continuous model generation, ensuring robust adaptation to dynamic resource constraints.
The rapid expansion of Internet of Things (IoT) devices has led to an explosion of event data, posing significant challenges for traditional process model discovery techniques in terms of scalability and discovery accuracy. These techniques rely on centralized storage and processing, which are hindered by data transfer limitations, storage capacity, and computational overhead in distributed IoT environments. Edge-based model discovery techniques offer a promising solution for analyzing large-scale IoT data. However, existing techniques suffer from low efficiency and an inability to handle complex process structures. To address these challenges, we propose EdgeIM, an efficient edge-based process model discovery technique that enhances efficiency and model accuracy. EdgeIM operates in three key stages: preprocessing and feature-preserving sampling to eliminate redundant data, local processing at edge nodes to extract key structural features, and global feature aggregation at a central node for model discovery. EdgeIM has been implemented on the open-source process mining platform PM4Py, and experimental results on nine public event logs demonstrate that, compared to existing edge-based model discovery techniques, EdgeIM significantly improves discovery efficiency while maintaining high model quality.
Predictive Process Monitoring (PPM) is a critical technology for analyzing log data to forecast future events in ongoing process traces, enabling early warnings of business risks and timely interventions. Traditional PPM methods primarily focus on predicting a subsequent activity based solely on activity sequences. However, such approaches often neglect the suffix variability reflected by branching behavior, where identical prefixes may lead to multiple potential outcomes. To address this limitation, we propose a two-stage PPM model based on enabled state filtering for coarse-to-fine activity prediction. In the first stage, the model identifies all enabled activities based on the activity prefix, enhancing flexibility to handle variable suffixes. In the second stage, the model further integrates multi-attribute trace information along with the candidate activity set, employing a context-aware attention mechanism tailored to these candidates for fine-grained prediction. The constraints of the first stage allow the second stage to focus on the possible activities and relevant attributes, effectively reducing the search space and mitigating overfitting. Experimental results on four real-world datasets show that our approach outperforms traditional methods, particularly with variable suffixes. Ablation studies further highlight the critical role of the enabled state filtering strategy in reducing overfitting when modeling multi-attribute information.
Edge service deployment has attracted significant attention in recent years, aiming to optimize service placement on edge servers while satisfying diverse requirements. However, existing approaches often overlook the influence of geographical contexts on service demands, where user needs vary significantly across regions with distinct characteristics. They also fail to account for differences between direct responses and multi-hop forwarding in edge network topology, leading to unsatisfactory edge service deployment strategies. To this end, we formulate the Points of Interest-Based Edge Service Deployment (POIESD) problem with topology-aware optimization, integrating POI attributes and spatial distributions while incorporating edge network topology to enhance service placement. By proving the NP-hardness of POI-ESD problem, we propose a novel graph-encoded genetic algorithm, MTGA, to efficiently generate highquality deployment strategies. It ensures strategic placement of edge services in regions that best match user demands, improving the service utilization and satisfiability for edge users. Extensive experiments on a real-world dataset combining Shanghai Telecom and Baidu Maps POI data demonstrate that MTGA significantly outperforms existing competing approaches, achieving superior performance of edge service deployment.
Autoregressive transformer models have achieved state-of-the-art performance in advanced services such as text generation and machine translation. Given the significant computational bottlenecks of model inference, layer-wise skipping has emerged as a promising method to accelerate inference by bypassing redundant layers. However, existing methods face challenges, including sub-optimal performance resulting from the premature skipping of critical layers and an unbalanced focus on either multi-head attention or feed-forward sub-blocks, ultimately leading to global performance degradation. In light of the above challenges, we propose a Dynamic Inference Method, named DIM, for autoregressive transformer models. DIM dy-namically selects sub-blocks from both multi-head attention and feed-forward networks through the importance score alignment, ensuring a balanced selection that optimizes both efficiency and model performance. To further mitigate the potential performance loss of skipped sub-blocks, a lightweight adjustment is developed to approximate the computations of skipped sub-blocks. Finally, extensive experiments using several benchmarks validate that DIM outperforms existing inference methods.
We propose a new workflow task execution time prediction approach, DAG-FGL, by integrating Flash attention mechanism with a GraphLSTM model. It addresses the challenge of low task execution time prediction accuracy in the presence of complex dependencies among workflow subtasks. The GraphLSTM model captures and conveys subtask dependencies through the adjacency matrix of sub task relationships modeled as a directed acyclic graph (DAG). The Flash attention mechanism enhances the model by incorporating customized positional encoding of subtask priority. The encoding ensures that the model accurately reflects each subtask's importance and relative order when calculating attention weight. Therefore, DAG- FG L can more accurately predict task execution time in the context of complex dependencies. Experimental results show that DAG-FGL outperforms the best-performing baseline model. It achieves 7.82 % to 44.52 % improvements in prediction accuracies over the best-performing baselines across three cloud workflow datasets of varying lengths.
Unmanned aerial vehicle (UAV) swarms are useful for mobile and collaborative applications due to their flexibility, scalability, and reliability. However, managing their communication and collaboration in complex environments is challenging. Service mesh has demonstrated excellent performance in managing communication between microservices in cloudnative environments. However, its centralized and static network structure design hinders its adaptability to dynamic topologies, increases vulnerability to single points of failure, and exacerbates resource constraints when applied to UAV swarms. To address these challenges, we propose UAV-Mesh, a graph-based decentralized service mesh framework for UAV swarms. It models the swarm as a dynamic graph for enabling the data plane to adapt to changing topologies, mitigates the risk of single points of failure through a decentralized control plane, and addresses resource constraints by optimizing consensus mechanism and algorithm. UAV Mesh offers a decentralized perspective for the application of service mesh in UAV swarms. Through experiments and analysis involving varying numbers of UAVs in a complex scenario, we demonstrate the effectiveness and efficiency of UAV Mesh in managing and controlling UAV services.
As application migration to the cloud becomes the mainstream way of application deployment, accurate workload prediction is critical to ensure the quality of service (QoS) and cost-efficiency of the applications and meet service level agreements (SLAs) with users. Short-term workload prediction can handle workload fluctuations over a short duration, while long-term prediction can capture trend and periodic changes for workload. However, existing studies are unable to deal with long-term forecasts in multiple scales effectively; moreover, new containers lack historical data, leading to inaccurate prediction, while existing methods perform poorly due to a lack of generalization ability. To tackle these problems, we propose LLM4Load, an LLM Prompt-Driven Framework for Multi-Scale Workload Prediction. Firstly, we design a structured prompt with dataset introduction, task description, and workload features characterization to extract multi-scale features. Secondly, we introduce a cross-modality alignment mechanism combined with label embedding to further enhance performance. Leveraging the generalization capability of LLM, we also solve the container cold-start problem. The abundant experiments on four real cloud workload datasets validate the effectiveness of LLM4Load. For multi-scale workload prediction, LLM4Load improves up to 42.40%. For container cold-start scenarios, LLM4Load improves up to 93.72%. These results highlight the potential of LLM4Load to drive dynamic and efficient resource management in modern cloud systems.
GraphQL offers a flexible alternative to REST APIs, enabling precise data retrieval across multiple services—a critical requirement in today's service-oriented architectures. However, constructing complex GraphQL queries remains challenging, and even Large Language Models (LLMs) often generate suboptimal queries due to limited schema awareness. Recent advancements, such as specialized prompt engineering, schema-aware in-context learning, and dedicated datasets, have aimed to improve query generation. However, evaluating the quality of generated queries remains challenging: GraphQL's inherent flexibility allows semantically equivalent queries to differ syntactically, complicating both automatic and manual evaluation. In this work, we introduce Robust GraphQL Evaluation (RGEval), the first benchmarking pipeline designed to systematically assess the quality of LLM-generated GraphQL queries. RGEval effectively handles schema complexities and structural variations, ensuring accurate evaluations while significantly improving efficiency—reducing evaluation time from hours to minutes. With Gartner projecting that over 60 % of enterprises will adopt GraphQL in production by 2027, RGEval provides a critical solution for benchmarking LLM-generated queries, fostering trust in AI-driven web service consumption.
Centralized storage in web-based services poses privacy risks for diverse user data, whereas personalized federated learning (PFL) arnesses the inherent privacy-preserving benefits of FL through distributed collaborative training, avoiding the centralization of sensitive data. Besides, PFL also customizes models to individual client needs, enabling personalized web services on non-independent and identically distributed data across varied environments. Existing PFL techniques often rely on prior knowledge to identify personalization layers, overlooking variable layer sensitivities to heterogeneous data, thus constraining the model's adaptability to diverse statistical distributions and limiting its service effectiveness. To overcome this issue, we introduce FedLD, a layer-wise decoupling method for PFL that precisely captures layer-specific sensitivities to heterogeneous data, enhancing service-oriented model personalization. Firstly, FedLD employs a hypernetwork to evaluate each layer's contribution to performance across varied data distributions, determining the optimal proportion of personalized channels per layer for each client, thus enabling fine-grained, channel-level parameter adjustments tailored to individual service needs. Secondly, knowledge distillation is introduced to enforce consistency between the representations of personalized and shared weights, promoting collaboration between them. Thirdly, an improved aggregation strategy is proposed to enhance collaboration among clients with similar data distributions. Extensive experimental results show that FedLD achieves more effective and generalized results compared to baseline methods across various types of heterogeneous data settings.
Using pruning techniques to prune redundant weights in large language models (LLMs) for model size reduction, enabling deployment on devices to deliver high-quality services, has garnered significant attention from industry and academia. However, most existing pruning methods suffer from two major problems: (1) they rely on operations with high computational complexity, such as Hessian matrix calculation, and (2) high pruning rate leads to a significant decrease in model accuracy. To this end, this paper proposes a low resource requirement and low accuracy loss post-training pruning approach, namely the compensated post-training pruning method (CPP) for on-device LLM services. First, CPP employs singular value decomposition on weight matrices, sorting the decomposed singular values in descending order. Based on the pruning ratio, it retains the principal eigenvector corresponding to the larger singular values. To maintain consistent output feature distributions, CPP applies orthogonal transformations to input data and weight matrices, leveraging the principle of matrix orthogonality invariance. Compared with pruning approaches based on the Hessian matrix, CPP does not need to iteratively calculate the second derivative, thereby avoiding a high computational overhead. Second, the CPP incorporates bias compensation to use valuable information in pruned weights to improve the accuracy of the model. It constructs mapping relationships between input features and pruned weights through tensor decomposition techniques to generate bias compensation terms. These terms fine-tune the output of each layer, reducing pruning-induced errors from ratio-based pruning, and consequently improving model accuracy. Finally, CPP is validated through comprehensive experiments on nine benchmark datasets (e.g., WikiText-2 and PIQA) using large language models (i.e., LLaMA and OPT). Specifically, compared with the FLAP, CPP demonstrates 6.68% reduction in average pruning time. At pruning ratios of 10% and 20%, CPP achieves 2. 14% and 1.45% reductions in perplexity, respectively, along with 1.22% to 3.43% improvement in zeroshot inference accuracy.
Permissioned blockchains have found widespread adoption across diverse scenarios, ensuring data authenticity and integrity. However, transaction conflicts, as an inherent performance challenge in permissioned blockchains, can significantly decrease system throughput and thus degrade its Quality of Service (QoS) under substantial transaction contention. Existing approaches mitigate conflicts typically by either aborting or blocking transactions in advance, encountering two main issues: (i) resource wastage due to transaction failure and (ii) performance degradation, particularly under large block sizes or high transaction contention. In this paper, we propose PreFabric, a novel permissioned blockchain framework that guarantees high throughput by resolving the transaction conflict problem. We first conduct a comprehensive analysis of the transaction scenarios preceding simulation execution of the endorsing phase in the blockchain system to identify potential conflict-causing situations. Then, we devise an key-locking method to prevent transaction conflicts and propose concurrency control strategies based on dependency analysis, encompassing a transaction merging mechanism, an key-renaming mechanism and concurrent validating mechanisms, to improve system throughput. The experimental results demonstrate the superior performance of our method over state-of-the-art methods, with 2.1x higher effective throughput and 0.48x lower latency.
Regression testing enhances software reliability by detecting regressions in new versions. Regression test suites often lack awareness of real-world product/service usage, potentially leading to undetected faults and ineffective testing scenarios. We propose LogTest, a transformer-based approach that learns from event logs and system traces to automatically generate service invocation sequences that mimic observed system behaviors. These sequences enhance regression test suites by exposing past real execution patterns. A preliminary experimentation on a realistic benchmark demonstrates its potential.
Financial trading analytics increasingly demands modular, explainable, and adaptive intelligence systems capable of handling volatile market conditions and multimodal data streams. Recent advances in generative Artificial Intelligence (AI), Large Language Models (LLMs), and graph-based representations have enabled the creation of intelligent agents that can reason over complex, interconnected financial data. We introduce GWise, a graph-structured, generative AI-enabled multi-agent framework for real-time financial trading analytics delivered via secure web services. GWise models financial decision making as a directed computational graph of specialized analytical agents, including technical, fundamental, sentiment, and risk analysis crews, whose outputs are orchestrated through a memory-augmented LLM. This graph-structured design enables transparent, adaptive, and explainable trade recommendations that evolve over time. We demonstrate how the framework's agent orchestration forms a dynamic service graph, facilitating composability, fault isolation, and scalable deployment through cloud-native APIs. Extensive back-testing and simulated market conditions show that GWise outperforms traditional strategies in risk-adjusted returns while offering improved interpretability and service robustness. Our work illustrates how graph-based multiagent coordination and generative reasoning can advance real-time financial analytics as a service.
