Balancing Act: Constraining Disparate Impact in Sparse Models

arXiv:2310.20673v2 Announce Type: replace
Abstract: Model pruning is a popular approach to enable the deployment of large deep learning models on edge devices with restricted computational or storage capacities. Although sparse models achieve performance comparable to that of their dense counterparts at the level of the entire dataset, they exhibit high accuracy drops for some data sub-groups. Existing methods to mitigate this disparate impact induced by pruning (i) rely on surrogate metrics that address the problem indirectly and have limited …

abstract accuracy act arxiv balancing act computational cs.cy cs.lg dataset deep learning deployment devices edge edge devices impact performance popular pruning storage type