Abstract:Sometimes local search algorithms cannot efficiently find even local peaks. To understand why, I look at the structure of ascents in fitness landscapes from valued constraint satisfaction problems (VCSPs) parameterized by the treedepth of their constraint graphs. There are existing constructions of VCSPs with logarithm treedepth that represent fitness landscapes where all ascents are exponential from some initial assignment. I improve these bounds by showing that with loglog treedepth, superpolynomial ascents exist; and for polylog treedepth, there are initial assignments from which all ascents are superpolynomial. My hope is that these examples of sparse VCSPs can help us better understand the barriers to efficient local search.
From: Artem Kaznatcheev [view email]
[v1]
Mon, 20 May 2024 23:28:38 UTC (23 KB)
[v2]
Wed, 22 Jul 2026 13:54:48 UTC (8 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。