Towards blackbox identity testing of log-variate circuits

Towards blackbox identity testing of log-variate circuits
复制标题

DOI:
10.4230/lipics.icalp.2018.54
复制
发表时间:
2018-07
期刊:
--
影响因子:
--
通讯作者:
Michael A. Forbes;Sumanta K Ghosh;Nitin Saxena
Michael A. Forbes;Sumanta K Ghosh;Nitin Saxena
中科院分区:
其他
文献类型:
--
作者:
Michael A. Forbes;Sumanta K Ghosh;Nitin Saxena

文献摘要

相似文献

[12]黑盒身份测试的去随机化简化为极其特殊的电路模型。经过13一行的工作,我们知道,专注于电路与恒定的深度和不断许多14变量是不够的(阿格拉瓦尔,高希,Saxena,STOC'18),以获得一般击中集和电路15下界。这启发我们去研究变量少的电路,例如。在大小s上的对数。[16]我们给出了n = O(log s)变量大小电路17的第一个poly(s)时间黑盒恒等式检验,该电路具有poly(s)维偏导数空间;例如。深度为3的对角线电路(或称对角线电路)。[18]前一个模型已经得到了很好的研究(Nisan,Wigderson,FOCS'95),但是在我们之前还没有poly(s 2 n)-时间恒等式[19]检验。本文介绍了锥闭基隔离的概念,并证明了它在研究对数变量电路中的有用性。它包含了先前在ROABP模型背景下广泛研究的秩-浓度21概念。22
12 Derandomization of blackbox identity testing reduces to extremely special circuit models. After 13 a line of work, it is known that focusing on circuits with constant-depth and constantly many 14 variables is enough (Agrawal,Ghosh,Saxena, STOC’18) to get to general hitting-sets and circuit 15 lower bounds. This inspires us to study circuits with few variables, eg. logarithmic in the size s . 16 We give the first poly( s )-time blackbox identity test for n = O (log s ) variate sizes circuits 17 that have poly( s )-dimensional partial derivative space; eg. depth-3 diagonal circuits (or Σ ∧ Σ n ). 18 The former model is well-studied (Nisan,Wigderson, FOCS’95) but no poly( s 2 n )-time identity 19 test was known before us. We introduce the concept of cone-closed basis isolation and prove its 20 usefulness in studying log-variate circuits. It subsumes the previous notions of rank-concentration 21 studied extensively in the context of ROABP models. 22