On the Existence of Universal Lottery TicketsDownload PDF

29 Sept 2021, 00:32 (edited 16 Mar 2022)ICLR 2022 PosterReaders: Everyone
  • Keywords: theory, deep learning, lottery tickets, universality
  • Abstract: The lottery ticket hypothesis conjectures the existence of sparse subnetworks of large randomly initialized deep neural networks that can be successfully trained in isolation. Recent work has experimentally observed that some of these tickets can be practically reused across a variety of tasks, hinting at some form of universality. We formalize this concept and theoretically prove that not only do such universal tickets exist but they also do not require further training. Our proofs introduce a couple of technical innovations related to pruning for strong lottery tickets, including extensions of subset sum results and a strategy to leverage higher amounts of depth. Our explicit sparse constructions of universal function families might be of independent interest, as they highlight representational benefits induced by univariate convolutional architectures.
  • One-sentence Summary: We formalize a notion of strong universal lottery tickets and prove their existence as subnetworks of randomly initialized neural networks.
  • Supplementary Material: zip
13 Replies