Risk-Sensitive Task Fetching and Offloading for Vehicular Edge Computing

TL;DR

This paper introduces a risk-sensitive approach to optimize task fetching and offloading in vehicular edge computing, significantly reducing delay variability and tail risk through a distributed learning algorithm.

Contribution

It proposes a novel risk-sensitive framework and a distributed learning algorithm for reliable task offloading in vehicular edge networks, improving delay stability.

Findings

Achieves up to 40% variance reduction in delay

Reduces tail risk of end-to-end delay distribution

Outperforms baseline methods in simulations

Abstract

This letter studies an ultra-reliable low latency communication problem focusing on a vehicular edge computing network in which vehicles either fetch and synthesize images recorded by surveillance cameras or acquire the synthesized image from an edge computing server. The notion of risk-sensitive in financial mathematics is leveraged to define a reliability measure, and the studied problem is formulated as a risk minimization problem for each vehicle's end-to-end (E2E) task fetching and offloading delays. Specifically, by resorting to a joint utility and policy estimation-based learning algorithm, a distributed risk-sensitive solution for task fetching and offloading is proposed. Simulation results show that our proposed solution achieves performance improvements up to 40% variance reduction and steeper distribution tail of the E2E delay over an averaged-based baseline.