Mitigating GPU Fragmentation and Tail Delay in Edge Computing: A Liquid-Driven Graph Reinforcement Learning Algorithm

Xiao He

doi:10.1016/j.eswa.2026.134692

Mitigating GPU Fragmentation and Tail Delay in Edge Computing: A Liquid-Driven Graph Reinforcement Learning Algorithm

2026-10-10

0

PRE

AI

DOI：10.1016/j.eswa.2026.134692

Original

Original request for help

Save

Abstract

En 中文

• We propose a collaborative scheduling framework for resource optimization. • We present CVaR-Lyapunov methods to reduce GPU fragmentation and tail delay. • We design Liquid-GRL to enhance reinforcement learning based scheduling decision.

Keywords:

Edge computing

Recourse allocation

Graph neural network

Reinforcement learning

Liquid neural network

Mitigating GPU Fragmentation and Tail Delay in Edge Computing: A Liquid-Driven Graph Reinforcement Learning Algorithm

Abstract

Journal

Organization

Cited Papers