Feature Likelihood Divergence: Evaluating the Generalization of Generative Models Using Samples

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jiralerspong, Marco, Bose, Avishek Joey, Gemp, Ian, Qin, Chongli, Bachrach, Yoram, Gidel, Gauthier
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914711159701504
author Jiralerspong, Marco
Bose, Avishek Joey
Gemp, Ian
Qin, Chongli
Bachrach, Yoram
Gidel, Gauthier
author_facet Jiralerspong, Marco
Bose, Avishek Joey
Gemp, Ian
Qin, Chongli
Bachrach, Yoram
Gidel, Gauthier
contents The past few years have seen impressive progress in the development of deep generative models capable of producing high-dimensional, complex, and photo-realistic data. However, current methods for evaluating such models remain incomplete: standard likelihood-based metrics do not always apply and rarely correlate with perceptual fidelity, while sample-based metrics, such as FID, are insensitive to overfitting, i.e., inability to generalize beyond the training set. To address these limitations, we propose a new metric called the Feature Likelihood Divergence (FLD), a parametric sample-based metric that uses density estimation to provide a comprehensive trichotomic evaluation accounting for novelty (i.e., different from the training samples), fidelity, and diversity of generated samples. We empirically demonstrate the ability of FLD to identify overfitting problem cases, even when previously proposed metrics fail. We also extensively evaluate FLD on various image datasets and model classes, demonstrating its ability to match intuitions of previous metrics like FID while offering a more comprehensive evaluation of generative models. Code is available at https://github.com/marcojira/fld.
format Preprint
id arxiv_https___arxiv_org_abs_2302_04440
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Feature Likelihood Divergence: Evaluating the Generalization of Generative Models Using Samples
Jiralerspong, Marco
Bose, Avishek Joey
Gemp, Ian
Qin, Chongli
Bachrach, Yoram
Gidel, Gauthier
Machine Learning
Computer Vision and Pattern Recognition
The past few years have seen impressive progress in the development of deep generative models capable of producing high-dimensional, complex, and photo-realistic data. However, current methods for evaluating such models remain incomplete: standard likelihood-based metrics do not always apply and rarely correlate with perceptual fidelity, while sample-based metrics, such as FID, are insensitive to overfitting, i.e., inability to generalize beyond the training set. To address these limitations, we propose a new metric called the Feature Likelihood Divergence (FLD), a parametric sample-based metric that uses density estimation to provide a comprehensive trichotomic evaluation accounting for novelty (i.e., different from the training samples), fidelity, and diversity of generated samples. We empirically demonstrate the ability of FLD to identify overfitting problem cases, even when previously proposed metrics fail. We also extensively evaluate FLD on various image datasets and model classes, demonstrating its ability to match intuitions of previous metrics like FID while offering a more comprehensive evaluation of generative models. Code is available at https://github.com/marcojira/fld.
title Feature Likelihood Divergence: Evaluating the Generalization of Generative Models Using Samples
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2302.04440