arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.02482stat.MLcs.LGmath.STstat.TH
收藏

抽样的费舍尔信息下界

Fisher information lower bounds for sampling

MIT(麻省理工学院);Johns Hopkins University(约翰斯·霍普金斯大学);Microsoft Research(微软研究院)

Sinho Chewi, Patrik Gerber, Holden Lee, Chen Lu

摘要:

我们在Balasubramanian et al.(2022)的框架内证明了非对数凹采样复杂性的两个下界,Balasubramanian et al.(2022)引入了使用费舍尔信息(FI)界作为采样中近似一阶平稳性的概念。我们的第一个下界表明,平均LMC是最佳的制度,大FI减少的问题,找到固定点在非凸优化采样。我们的第二个下界表明,在小FI的情况下,从目标分布获得至多$\varepsilon^2$的FI需要$\text{poly}(1/\varepsilon)$查询,这是令人惊讶的,因为它排除了高准确度算法的存在(例如,使用Metropolis-Hastings滤波器的算法)。

英文摘要:

We prove two lower bounds for the complexity of non-log-concave sampling within the framework of Balasubramanian et al. (2022), who introduced the use of Fisher information (FI) bounds as a notion of approximate first-order stationarity in sampling. Our first lower bound shows that averaged LMC is optimal for the regime of large FI by reducing the problem of finding stationary points in non-convex optimization to sampling. Our second lower bound shows that in the regime of small FI, obtaining a FI of at most $\varepsilon^2$ from the target distribution requires $\text{poly}(1/\varepsilon)$ queries, which is surprising as it rules out the existence of high-accuracy algorithms (e.g., algorithms using Metropolis-Hastings filters) in this context.

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。

↑