arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2107.00488cs.AIcs.LG
收藏

基于条件归一化流的可微粒子滤波器

Differentiable Particle Filters through Conditional Normalizing Flow

University of Surrey(萨里大学)

Xiongjie Chen, Hao Wen, Yunpeng Li

摘要:

微分粒子滤波器提供了一种灵活的机制,通过从观测数据中学习,自适应地训练动态和测量模型。然而,大多数现有的可微粒子滤波器是在自举粒子滤波框架内,并未能纳入最新的观测信息,以构建更好的建议。在本文中,我们利用条件规范化流构造建议分布的可微粒子滤波器,丰富的分布族,建议分布可以表示。此外,规范化流被纳入动态模型的构造中,从而产生更具表达力的动态模型。我们证明了所提出的条件归一化流为基础的可微粒子滤波器在视觉跟踪任务的性能。

英文摘要:

Differentiable particle filters provide a flexible mechanism to adaptively train dynamic and measurement models by learning from observed data. However, most existing differentiable particle filters are within the bootstrap particle filtering framework and fail to incorporate the information from latest observations to construct better proposals. In this paper, we utilize conditional normalizing flows to construct proposal distributions for differentiable particle filters, enriching the distribution families that the proposal distributions can represent. In addition, normalizing flows are incorporated in the construction of the dynamic model, resulting in a more expressive dynamic model. We demonstrate the performance of the proposed conditional normalizing flow-based differentiable particle filters in a visual tracking task.

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。

↑