2020/09/18 by Jonathan Frankle, Gintare Karolina Dziugaite, Frankle, Jonathan +5 · 2 voices · 5 citations
Computer Science · Mathematics · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Machine Learning and Data Classification #cs.CV #cs.LG #cs.NE #stat.ML
paper · pdf · doi:10.48550/arxiv.2009.08576
arxiv published 2020/09/18 · arxiv updated 2021/03/21
Recent work has explored the possibility of pruning neural networks at initialization. We assess proposals for doing so: SNIP (Lee et al., 2019), GraSP (Wang et al., 2020), SynFlow (Tanaka et al., 2020), and magnitude pruning. Although these methods surpass the trivial baseline of random pruning, they remain below the accuracy of magnitude pruning after training, and we endeavor to understand why. We show that, unlike pruning after training, randomly shuffling the weights these methods prune within each layer or sampling new initial values preserves or improves accuracy. As such, the per-weight pruning decisions made by these methods can be replaced by a per-layer choice of the fraction of weights to prune. This property suggests broader challenges with the underlying pruning heuristics, the desire to prune at initialization, or both.