arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2107.02306cs.LGcs.CV
收藏

连接性至关重要:通过有效稀疏性视角看神经网络剪枝

Connectivity Matters: Neural Network Pruning Through the Lens of Effective Sparsity

Center for Data Science(数据科学中心);New York University(纽约大学);Courant Institute for Mathematical Sciences(库朗数学科学研究所)

Artem Vysogorets, Julia Kempe

AI总结:

本文提出有效稀疏性概念以纠正传统直接稀疏性对子网络连接失活的低估,基于此重新评估多种剪枝算法,并利用物理学压力分布类比设计出超越现有基线的逐层稀疏配额。

摘要:

神经网络剪枝是一个富有成果的研究领域,在高稀疏性场景下引起了越来越多的关注。该领域的基准测试严重依赖于对子网络稀疏性的忠实表示,传统上这被计算为被移除连接的比例(直接稀疏性)。然而,这一定义未能识别那些与底层子网络的输入或输出层断开连接的未剪枝参数,从而可能低估了实际的有效稀疏性:即失活连接的比例。虽然这种影响对于中等程度剪枝的网络(压缩率高达10-100倍)可能可以忽略不计,但我们发现它对于更稀疏的子网络起着越来越大的作用,极大地扭曲了不同剪枝算法之间的比较。例如,我们展示了随机剪枝的 LeNet-300-100 的有效压缩率可能比其直接压缩率高出几个数量级,而使用 SynFlow 进行剪枝 [Tanaka et al., 2020] 时则从未观察到这种差异。在这项工作中,我们采用有效稀疏性的视角,在常见基准架构(例如 LeNet-300-100、VGG-19、ResNet-18)上重新评估了几种最近的剪枝算法,并发现它们在这个新的、更合适的框架中的绝对和相对性能发生了巨大变化。为了以有效稀疏性而非直接稀疏性为目标,我们为大多数剪枝算法开发了一种低成本的扩展。此外,在以有效稀疏性作为参考框架的情况下,我们部分地重新确认了:在初始化剪枝时,具有适当跨层稀疏性分配的随机剪枝,其表现与更复杂的算法一样好甚至更好 [Su et al., 2020]。针对这一观察结果,利用物理学中耦合圆柱体内压力分布的简单类比,我们设计了新颖的逐层稀疏配额,在随机剪枝的背景下优于所有现有的基线。

英文摘要:

Neural network pruning is a fruitful area of research with surging interest in high sparsity regimes. Benchmarking in this domain heavily relies on faithful representation of the sparsity of subnetworks, which has been traditionally computed as the fraction of removed connections (direct sparsity). This definition, however, fails to recognize unpruned parameters that detached from input or output layers of underlying subnetworks, potentially underestimating actual effective sparsity: the fraction of inactivated connections. While this effect might be negligible for moderately pruned networks (up to 10-100 compression rates), we find that it plays an increasing role for thinner subnetworks, greatly distorting comparison between different pruning algorithms. For example, we show that effective compression of a randomly pruned LeNet-300-100 can be orders of magnitude larger than its direct counterpart, while no discrepancy is ever observed when using SynFlow for pruning [Tanaka et al., 2020]. In this work, we adopt the lens of effective sparsity to reevaluate several recent pruning algorithms on common benchmark architectures (e.g., LeNet-300-100, VGG-19, ResNet-18) and discover that their absolute and relative performance changes dramatically in this new and more appropriate framework. To aim for effective, rather than direct, sparsity, we develop a low-cost extension to most pruning algorithms. Further, equipped with effective sparsity as a reference frame, we partially reconfirm that random pruning with appropriate sparsity allocation across layers performs as well or better than more sophisticated algorithms for pruning at initialization [Su et al., 2020]. In response to this observation, using a simple analogy of pressure distribution in coupled cylinders from physics, we design novel layerwise sparsity quotas that outperform all existing baselines in the context of random pruning.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑