Principled Interpolation in Normalizing Flows

Autor(es)

Fadel, Samuel G. ; Mair, Sebastian ; Torres, Ricardo da S. ; Brefeld, Ulf ; Oliver, N ; PerezCruz, F ; Kramer, S ; Read, J ; Lozano, JA

• Tipo de documento: Artigo Científico
• Fonte: MACHINE LEARNING AND KNOWLEDGE DISCOVERY IN DATABASES, ECML PKDD 2021: RESEARCH TRACK, PT II; v. 12976, p. 16-pg., 2021-01-01.

Resumo

Generative models based on normalizing flows are very successful in modeling complex data distributions using simpler ones. However, straightforward linear interpolations show unexpected side effects, as interpolation paths lie outside the area where samples are observed. This is caused by the standard choice of Gaussian base distributions and can be seen in the norms of the interpolated samples as they are outside the data manifold. This observation suggests that changing the way of interpolating should generally result in better interpolations, but it is not clear how to do that in an unambiguous way. In this paper, we solve this issue by enforcing a specific manifold and, hence, change the base distribution, to allow for a principled way of interpolation. Specifically, we use the Dirichlet and von Mises-Fisher base distributions on the probability simplex and the hypersphere, respectively. Our experimental results show superior performance in terms of bits per dimension, Frechet Inception Distance (FID), and Kernel Inception Distance (KID) scores for interpolation, while maintaining the generative performance. (AU)

• Processo FAPESP: 17/24005-2 - Inferência relacional temporal com redes neurais
• Beneficiário: Samuel Gomes Fadel
• Modalidade de apoio: Bolsas no Brasil - Doutorado
• Processo FAPESP: 18/19350-5 - Redes neurais para inferência relacional temporal na análise de futebol
• Beneficiário: Samuel Gomes Fadel
• Modalidade de apoio: Bolsas no Exterior - Estágio de Pesquisa - Doutorado