Dora: QoE-Aware Hybrid Parallelism for Distributed Edge AI

This paper introduces Dora, a framework for Quality of Experience (QoE) aware hybrid parallelism in distributed edge AI training and inference. It addresses the challenge of optimizing heterogeneous computation, contention-prone networks, and multi-dimensional QoE objectives in resource-constrained edge environments. Dora achieves this through a heterogeneity-aware model partitioner, a contention-aware network scheduler, and a runtime adapter, resulting in faster execution and reduced energy consumption while maintaining QoE.