Researchers have developed a novel framework for managing energy consumption and scheduling artificial intelligence-generated content (AIGC) workloads in distributed data centers. The approach addresses challenges like model heterogeneity and complex inference processes by characterizing service quality and optimizing system utility. To overcome the reward sparsity issue in deep reinforcement learning for scheduling, a diffusion model-aided reward shaping technique is introduced to synthesize reward signals, leading to more efficient learning and improved system performance. AI
影响 This research could lead to more efficient data center operations for AI content generation, potentially lowering costs and improving service quality.
排序理由 This is a research paper published on arXiv detailing a new technical approach.
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
