arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2004.03553cs.LGstat.ML
收藏

胶囊网络--一个概率的视角

Capsule Networks -- A Probabilistic Perspective

University of Oxford(牛津大学);Imperial College London(帝国理工学院)

Lewis Smith, Lisa Schut, Yarin Gal, Mark van der Wilk

摘要:

“胶囊”模型试图明确地表示对象的姿态,在对象的姿态和其组成部分之间强制执行线性关系。这种建模假设应该导致对视点变化的鲁棒性,因为子对象/超级对象关系对于对象的姿态是不变的。我们描述了一个概率生成模型,编码这样的胶囊假设,明确分离的推理机制模型的生成部分。随着变分约束,我们探索的属性的生成模型独立的近似推理方案,并获得洞察失败的胶囊假设和推断摊销。我们的实验证明了我们的统一目标的适用性,并演示了使用测试时间优化来解决我们的模型中的摊销推理所固有的问题。

英文摘要:

'Capsule' models try to explicitly represent the poses of objects, enforcing a linear relationship between an object's pose and that of its constituent parts. This modelling assumption should lead to robustness to viewpoint changes since the sub-object/super-object relationships are invariant to the poses of the object. We describe a probabilistic generative model which encodes such capsule assumptions, clearly separating the generative parts of the model from the inference mechanisms. With a variational bound we explore the properties of the generative model independently of the approximate inference scheme, and gain insights into failures of the capsule assumptions and inference amortisation. We experimentally demonstrate the applicability of our unified objective, and demonstrate the use of test time optimisation to solve problems inherent to amortised inference in our model.

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。

↑