SYShiyu Yang

Shiyu Yang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Shiyu Yang has a total of 8 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Top Cited Papers
    Open Access

    LLM-Driven Adaptive Cloud Resource Scheduling: Bridging Reasoning Intelligence With Optimization Guarantees

    by Guanyu Ding, Shiyu Yang, Han Lin, Zifan Chen, Jie Si Yang

    Cloud resource scheduling presents a fundamental challenge in modern distributed computing, where heterogeneous workloads, complex task dependencies, and multi-objective optimization requirements exceed the capabilities of traditional rule-based and small-scale machine learning approaches. Existing schedulers struggle to dynamically adapt to evolving workload patterns while simultaneously satisfying Service Level Agreement (SLA) constraints, resource efficiency targets, and fairness policies. This paper introduces LLMSched, a novel Large Language Model (LLM)-driven adaptive scheduling framework that synergistically combines the contextual reasoning capabilities of foundation models with the execution guarantees of classical optimization techniques. Our approach transforms heterogeneous cluster states-including task dependency graphs, real-time resource utilization metrics, and SLA specifications-into a unified structured-textual representation that leverages LLMs' few-shot learning and causal reasoning abilities to generate intelligent scheduling candidates. These candidates are subsequently refined through a lightweight Integer Linear Programming (ILP) module that ensures feasibility and optimality under resource constraints. We evaluate LLMSched on Google's production cluster trace dataset, demonstrating significant improvements over state-of-the-art baselines: 23.7% reduction in average job completion time, 18.4% improvement in resource utilization efficiency, and 31.2% decrease in SLA violations across diverse workload scenarios. Extensive ablation studies validate the contributions of each architectural component, while robustness analysis under workload perturbations confirms the framework's practical viability. Our work establishes a new paradigm for intelligent resource management that bridges the gap between neural reasoning and algorithmic precision, opening avenues for LLM applications in systems optimization domains.


  • Top Cited Papers
    Open Access

    GART: Graph Neural Network-Based Adaptive and Robust Task Scheduler for Heterogeneous Distributed Computing

    by Shiyu Yang, Guanyu Ding, Zifan Chen, Jie Si Yang

    Modern distributed computing systems face significant challenges in achieving optimal resource utilization and maintaining performance stability when operating in heterogeneous cluster environments with dynamic workloads. Existing task scheduling approaches often rely on heuristic rules or static policies that fail to adapt to runtime variations such as node failures, load fluctuations, and heterogeneous resource capacities. To address these limitations, we propose GART (Graph-based Adaptive Robust Task scheduling), a novel framework that leverages graph neural networks (GNN) and reinforcement learning for intelligent task scheduling in distributed systems. GART models distributed task workflows as directed acyclic graphs (DAGs) and employs a specially designed GNN architecture to capture complex inter-task dependencies and resource requirements. The framework learns optimal scheduling policies through deep reinforcement learning, enabling adaptive decision-making that responds to dynamic system conditions. Our approach incorporates a robust scheduling mechanism that maintains high throughput and low latency even under adverse conditions including node failures and sudden load spikes. Extensive experiments on the Alibaba cluster trace dataset demonstrate that GART achieves 23.7% improvement in average job completion time, 31.4% reduction in resource imbalance, and 42.6% better resilience to node failures compared to state-of-the-art baseline schedulers including Shortest Job First, Tetrisched, and Decima. The results validate GART’s effectiveness in real-world heterogeneous distributed computing environments and its potential for deployment in production systems.


  • Open Access

    Causal-Aware Multimodal Transformer for Supply Chain Demand Forecasting: Integrating Text, Time Series, and Satellite Imagery

    by Ying Wang, Guanyu Ding, Ziyang Zeng, Shiyu Yang

    Supply chain demand forecasting faces unprecedented challenges due to the complex interplay of multiple information sources, including market dynamics reflected in textual data, historical consumption patterns, and macroeconomic indicators captured through satellite imagery. Traditional forecasting methods primarily rely on historical time series data, failing to leverage the rich contextual information available from heterogeneous data modalities. This paper proposes a novel Causal-Aware Multimodal Transformer (CAMT) framework that systematically integrates textual data from news and social media, time series of historical demand, and satellite imagery reflecting economic activities for enhanced demand forecasting accuracy. Our approach introduces three key innovations: (1) a cross-modal attention mechanism that enables effective fusion of text-temporal-visual representations through learnable inter-modal relationships, (2) a causal discovery-based feature importance evaluation method that identifies genuine causal relationships between different modalities and demand patterns while mitigating spurious correlations, and (3) a hierarchical multi-scale prediction framework that provides coherent forecasts across product, category, and regional levels. The framework is built upon a Vision-Language Transformer architecture with specialized encoders for each modality and a unified decoder for demand prediction. Extensive experiments on the M5 Forecasting Competition dataset, augmented with news articles, social media posts, and corresponding satellite imagery, demonstrate that CAMT achieves significant improvements over state-of-the-art baselines, with 12.3% reduction in Root Mean Square Error (RMSE) and 15.7% improvement in Mean Absolute Percentage Error (MAPE) compared to the best performing baseline. Ablation studies confirm the effectiveness of each proposed component, while causal analysis reveals that satellite-derived economic indicators contribute most significantly to long-term forecasting accuracy, whereas textual sentiment shows stronger predictive power for short-term demand fluctuations. The proposed framework offers practical implications for supply chain management by enabling more accurate demand forecasting that considers diverse external factors, ultimately supporting better inventory management and operational planning decisions.