Scholarlink

With the growing adoption of Artificial Intelligence (AI) across industries, energy usage and efficiency have emerged as critical concerns. Balancing performance and sustainability is essential from environmental, social, and economic perspectives. For industrial applications, the edge-cloud continuum provides a range of deployment options, from centralized cloud data centers to decentralized edge devices. This enables the strategic placement of AI applications to optimize performance, latency, resource utilization, and energy efficiency. This work focuses on the energy-aware placement of distributed AI applications as part of the broader ESCADE project. It is dedicated to improving energy efficiency in data centers and AI systems through neuromorphic computing. We propose a solution for placing AI applications and services across the edge-cloud continuum, consisting of diverse hardware platforms such as neuromorphic processing units (NPUs) and traditional graphics processing units (GPUs). The solution is applied in a real-world scenario: AI-based steel scrap classification for recycling. The results show the potential of strategic placement for reducing energy consumption and environmental impact in distributed AI systems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Energy-Aware Placement of AI Applications in an Edge-Cloud Continuum with Neuromorphic Chips for Steel Scrap Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Madhu Dasika,
  • Jia Lei Du,
  • Narges Mehran,
  • Stefan Farthofer-Oster,
  • Dina Barbian,
  • Andreas Hantsch,
  • Sirine Arfa,
  • Ulrike Faltings,
  • Michael Schäfer

摘要

With the growing adoption of Artificial Intelligence (AI) across industries, energy usage and efficiency have emerged as critical concerns. Balancing performance and sustainability is essential from environmental, social, and economic perspectives. For industrial applications, the edge-cloud continuum provides a range of deployment options, from centralized cloud data centers to decentralized edge devices. This enables the strategic placement of AI applications to optimize performance, latency, resource utilization, and energy efficiency. This work focuses on the energy-aware placement of distributed AI applications as part of the broader ESCADE project. It is dedicated to improving energy efficiency in data centers and AI systems through neuromorphic computing. We propose a solution for placing AI applications and services across the edge-cloud continuum, consisting of diverse hardware platforms such as neuromorphic processing units (NPUs) and traditional graphics processing units (GPUs). The solution is applied in a real-world scenario: AI-based steel scrap classification for recycling. The results show the potential of strategic placement for reducing energy consumption and environmental impact in distributed AI systems.