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 2101.07752cs.LGmath.AT
收藏

持久同调神经网络相似性的刻画与度量

Characterizing and Measuring the Similarity of Neural Networks with Persistent Homology

SEGITTUR(西班牙旅游技术与管理公司(SEGITTUR));Barcelona Supercomputing Center(巴塞罗那超级计算中心)

David Pérez-Fernández, Asier Gutiérrez-Fandiño, Jordi Armengol-Estapé, Marta Villegas

摘要:

表征神经网络的结构特性是至关重要的,但人们对其了解甚少,并且网络之间没有完善的相似性度量。在这项工作中,我们观察到,神经网络可以表示为抽象的单纯复形,并通过持久同源性(PH)使用其拓扑“指纹”进行分析。然后,我们描述了一个基于PH的表示提出的神经网络的特征和测量相似性。我们凭经验证明了这种表示在几个数据集中作为不同架构的描述符的有效性。这种基于拓扑数据分析的方法是更好地理解神经网络的一个步骤,并作为一个有用的相似性度量。

英文摘要:

Characterizing the structural properties of neural networks is crucial yet poorly understood, and there are no well-established similarity measures between networks. In this work, we observe that neural networks can be represented as abstract simplicial complex and analyzed using their topological 'fingerprints' via Persistent Homology (PH). We then describe a PH-based representation proposed for characterizing and measuring similarity of neural networks. We empirically show the effectiveness of this representation as a descriptor of different architectures in several datasets. This approach based on Topological Data Analysis is a step towards better understanding neural networks and serves as a useful similarity measure.

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。

↑