arXiv cs.AI
· Papers
Denial of Deadline: Network-Driven Accuracy Collapse in Distributed Inference Pipelines
arXiv:2607.24692v1 Announce Type: cross Abstract: Inference systems increasingly combine a fast path that returns predictions within the application's latency deadline together with a higher-accuracy slow path that runs higher-compute methods on stronger, remote hardware, so its results can be returned on time and comb