arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2201.11211stat.MLcs.LGcs.SYeess.SYmath.STstat.TH
收藏

线性动力系统的混合学习

Learning Mixtures of Linear Dynamical Systems

Princeton University(普林斯顿大学)

Yanxi Chen, H. Vincent Poor

摘要:

我们研究了学习混合的多个线性动力系统(LDS)从未标记的短样本轨迹,每个生成的LDS模型之一的问题。尽管时间序列数据的混合模型具有广泛的适用性,但现有文献中基本上没有端到端性能保证的学习算法。存在多个技术挑战来源,包括但不限于(1)潜在变量(即轨迹的未知标签)的存在;(2)样本轨迹可能具有比LDS模型的维度$d$小得多的长度的可能性;以及(3)时间序列数据固有的复杂时间依赖性。为了解决这些挑战,我们开发了一个两阶段的元算法,这是保证有效地恢复每个地面实况LDS模型的错误$\tilde{O}(\sqrt{d/T})$,其中$T$是总样本量。我们验证了我们的理论研究与数值实验,确认所提出的算法的有效性。

英文摘要:

We study the problem of learning a mixture of multiple linear dynamical systems (LDSs) from unlabeled short sample trajectories, each generated by one of the LDS models. Despite the wide applicability of mixture models for time-series data, learning algorithms that come with end-to-end performance guarantees are largely absent from existing literature. There are multiple sources of technical challenges, including but not limited to (1) the presence of latent variables (i.e. the unknown labels of trajectories); (2) the possibility that the sample trajectories might have lengths much smaller than the dimension $d$ of the LDS models; and (3) the complicated temporal dependence inherent to time-series data. To tackle these challenges, we develop a two-stage meta-algorithm, which is guaranteed to efficiently recover each ground-truth LDS model up to error $\tilde{O}(\sqrt{d/T})$, where $T$ is the total sample size. We validate our theoretical studies with numerical experiments, confirming the efficacy of the proposed algorithm.

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。

↑