更快的张量规范化
Faster Tensor Canonicalization
摘要:
Butler-Portugal算法用于获得关于槽对称性和伪索引重命名的张量表达式的标准形式,在具有高度对称性的某些情况下,在计算时间和存储器两者中遭受$O(n!)$爆炸。我们提出了一个修改后的算法,它在最常见的情况下--张量表达式与子集的指数是完全对称或完全反对称--在多项式时间。我们还提出了一个实现的标签重命名机制,改进后,原来的巴特勒-葡萄牙算法,从而提供了一个显着的速度增加的平均情况下,以及高度对称的特殊情况。最坏情况行为仍然是$O(n!)$,尽管它发生在更有限的情况下,不太可能出现在实际计算中。我们评论可能采取的策略,如果计算的性质应该使这些情况更有可能。
英文摘要:
The Butler-Portugal algorithm for obtaining the canonical form of a tensor expression with respect to slot symmetries and dummy-index renaming suffers, in certain cases with a high degree of symmetry, from $O(n!)$ explosion in both computation time and memory. We present a modified algorithm which alleviates this problem in the most common cases---tensor expressions with subsets of indices which are totally symmetric or totally antisymmetric---in polynomial time. We also present an implementation of the label-renaming mechanism which improves upon that of the original Butler-Portugal algorithm, thus providing a significant speed increase for the average case as well as the highly-symmetric special case. The worst-case behavior remains $O(n!)$, although it occurs in more limited situations unlikely to appear in actual computations. We comment on possible strategies to take if the nature of a computation should make these situations more likely.
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。