Show simple item record

dc.contributor.authorElfeki, Mohamed
dc.contributor.authorCouprie, Camille
dc.contributor.authorRivière, Morgane
dc.contributor.authorElhoseiny, Mohamed
dc.date.accessioned2020-04-27T00:52:38Z
dc.date.available2020-04-27T00:52:38Z
dc.date.issued2019-01-01
dc.date.submitted2018-11-30
dc.identifier.citationInternational Conference on Machine Learning 2019
dc.identifier.isbn9781510886988
dc.identifier.urihttp://hdl.handle.net/10754/662645
dc.description.abstractGenerative models have proven to be an outstanding tool for representing high-dimensional probability distributions and generating realistic looking images. An essential characteristic of generative models is their ability to produce multimodal outputs. However, while training, they are often susceptible to mode collapse, that is models are limited in mapping input noise to only a few modes of the true data distribution. In this work, we draw inspiration from Determinantal Point Process (DPP) to propose an unsupervised penalty loss that alleviates mode collapse while producing higher quality samples. DPP is an elegant probabilistic measure used to model negative correlations within a subset and hence quantify its diversity. We use DPP kernel to model the diversity in real data as well as in synthetic data. Then, we devise an objective term that encourages generator to synthesize data with a similar diversity to real data. In contrast to previous state-of-the-art generative models that tend to use additional trainable parameters or complex training paradigms, our method docs not change the original training scheme. Embedded in an adversarial training and variational autoencoder, our Generative DPP approach shows a consistent resistance to mode-collapse on a wide-variety of synthetic data and natural image datasets including MNIST, CIFAR10, and CelcbA, while outperforming state-of-the-art methods for data-efficiency, generation quality, and convergence-time whereas being 5.8x faster than its closest competitor.
dc.publisherInternational Machine Learning Society (IMLS)rasmussen@ptd.net
dc.relation.urlhttp://proceedings.mlr.press/v97/elfeki19a.html
dc.rightsArchived with thanks to International Machine Learning Society (IMLS)rasmussen@ptd.net
dc.titleGDPP: Learning Diverse Generations using Determinantal Point Processes
dc.typeConference Paper
dc.contributor.departmentComputer, Electrical and Mathematical Sciences and Engineering (CEMSE) Division
dc.conference.date2019-06-09 to 2019-06-15
dc.conference.name36th International Conference on Machine Learning, ICML 2019
dc.conference.locationLong Beach, CA, USA
dc.eprint.versionPre-print
dc.contributor.institutionUniversity of Central Florida, United States
dc.contributor.institutionFacebook Artificial Intelligence Research
dc.identifier.volume2019-June
dc.identifier.pages3178-3193
dc.identifier.arxivid1812.00068
kaust.personElhoseiny, Mohamed
dc.date.accepted2019
dc.relation.issupplementedbygithub:M-Elfeki/GDPP
dc.identifier.eid2-s2.0-85079438441
refterms.dateFOA2020-04-27T00:56:10Z
display.relations<b>Is Supplemented By:</b><br/> <ul><li><i>[Software]</i> <br/> Title: M-Elfeki/GDPP: Generator loss to reduce mode-collapse and to improve the generated samples quality.. Publication Date: 2018-11-30. github: <a href="https://github.com/M-Elfeki/GDPP" >M-Elfeki/GDPP</a> Handle: <a href="http://hdl.handle.net/10754/667580" >10754/667580</a></a></li></ul>


Files in this item

Thumbnail
Name:
elfeki19a.pdf
Size:
1.113Mb
Format:
PDF
Description:
Conference Paper
Thumbnail
Name:
elfeki19a-supp.pdf
Size:
1.469Mb
Format:
PDF
Description:
Supplemental Material

This item appears in the following Collection(s)

Show simple item record