{"id":{"repo_id":"uiuc","oai_identifier":"oai:www.ideals.illinois.edu:2142/101060"},"canonical_url":"https://search.dev.ndltd.org/etd/uiuc/oai:www.ideals.illinois.edu:2142/101060","repository":{"repo_id":"uiuc","name":"University of Illinois - Urbana-Champaign","base_url":"https://www.ideals.illinois.edu/oai-pmh"},"display":{"title":"PacGAN: The power of two samples in generative adversarial networks","abstract":"Generative adversarial networks (GANs) are innovative techniques for learning generative models of complex data distributions from samples. Despite remarkable recent improvements in generating realistic images, one of their major shortcomings is the fact that in practice, they tend to produce samples with little diversity, even when trained on diverse datasets. This phenomenon, known as mode collapse, has been the main focus of several recent advances in GANs. Yet there is little understanding of why mode collapse happens and why existing approaches are able to mitigate mode collapse. We propose a principled approach to handling mode collapse, which we call {\\em packing}. The main idea is to modify the discriminator to make decisions based on multiple samples from the same class, either real or artificially generated. We borrow analysis tools from binary hypothesis testing---in particular the seminal result of Blackwell \\cite{Bla53}---to prove a fundamental connection between packing and mode collapse. We show that packing naturally penalizes generators with mode collapse, thereby favoring generator distributions with less mode collapse during the training process. Numerical experiments on benchmark datasets suggests that packing provides significant improvements in practice as well.","abstract_html":"Generative adversarial networks (GANs) are innovative techniques for learning generative models of complex data distributions from samples. Despite remarkable recent improvements in generating realistic images, one of their major shortcomings is the fact that in practice, they tend to produce samples with little diversity, even when trained on diverse datasets. This phenomenon, known as mode collapse, has been the main focus of several recent advances in GANs. Yet there is little understanding of why mode collapse happens and why existing approaches are able to mitigate mode collapse. We propose a principled approach to handling mode collapse, which we call {\\em packing}. The main idea is to modify the discriminator to make decisions based on multiple samples from the same class, either real or artificially generated. We borrow analysis tools from binary hypothesis testing---in particular the seminal result of Blackwell \\cite{Bla53}---to prove a fundamental connection between packing and mode collapse. We show that packing naturally penalizes generators with mode collapse, thereby favoring generator distributions with less mode collapse during the training process. Numerical experiments on benchmark datasets suggests that packing provides significant improvements in practice as well.","abstract_has_math":false,"creators":["Khetan, Ashish Kumar"],"institution":"University of Illinois at Urbana-Champaign","degree_name":"M.S.","degree_level":"Thesis","degree_discipline":"Computer Science","degree_department":null,"school":null,"contributors":["Oh, Sewoong","Koyejo, Sanmi"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2018,"date_issued":"2018-09-04T20:27:27Z","date_published":"2018-09-04T20:27:27Z","updated_at":"2026-07-22T22:24:38Z","subjects":["GAN"],"languages":["en"],"rights":["Copyright 2018 Ashish Kumar Khetan"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"http://hdl.handle.net/2142/101060","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Oh, Sewoong","Koyejo, Sanmi"]},{"key":"dc:creator","label":"Author","values":["Khetan, Ashish Kumar"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2018-09-04T20:27:27Z","2018-04-25","2018-05"]},{"key":"dc:type","label":"Dc Type","values":["text"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["M.S."]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Illinois at Urbana-Champaign"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["GAN"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Copyright 2018 Ashish Kumar Khetan"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["http://hdl.handle.net/2142/101060"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Generative adversarial networks (GANs) are innovative techniques for learning generative models of complex data distributions from samples. Despite remarkable recent improvements in generating realistic images, one of their major shortcomings is the fact that in practice, they tend to produce samples with little diversity, even when trained on diverse datasets. This phenomenon, known as mode collapse, has been the main focus of several recent advances in GANs. Yet there is little understanding of why mode collapse happens and why existing approaches are able to mitigate mode collapse. We propose a principled approach to handling mode collapse, which we call {\\em packing}. The main idea is to modify the discriminator to make decisions based on multiple samples from the same class, either real or artificially generated. We borrow analysis tools from binary hypothesis testing---in particular the seminal result of Blackwell \\cite{Bla53}---to prove a fundamental connection between packing and mode collapse. We show that packing naturally penalizes generators with mode collapse, thereby favoring generator distributions with less mode collapse during the training process. Numerical experiments on benchmark datasets suggests that packing provides significant improvements in practice as well.","Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2018-08-31 without embargo terms","The student, Ashish Kumar Khetan, accepted the attached license on 2018-04-25 at 10:58.","The student, Ashish Kumar Khetan, submitted this Thesis for approval on 2018-04-25 at 11:06.","This Thesis was approved for publication on 2018-04-25 at 13:42.","DSpace SAF Submission Ingestion Package generated from Vireo submission #12456 on 2018-08-31 at 17:14:33","Made available in DSpace on 2018-09-04T20:27:27Z (GMT). No. of bitstreams: 2 KHETAN-THESIS-2018.pdf: 864360 bytes, checksum: 7bd9e0dc7d28b7ce9be1a306630638a9 (MD5) LICENSE.txt: 4216 bytes, checksum: 3a692fbeb8fb8f2ed797c12c22c3e244 (MD5) Previous issue date: 2018-04-25"]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["PacGAN: The power of two samples in generative adversarial networks"]}]}],"canonical_facts":{"dc:contributor":["Oh, Sewoong","Koyejo, Sanmi"],"dc:creator":["Khetan, Ashish Kumar"],"dc:date":["2018-09-04T20:27:27Z","2018-04-25","2018-05"],"dc:description":["Generative adversarial networks (GANs) are innovative techniques for learning generative models of complex data distributions from samples. Despite remarkable recent improvements in generating realistic images, one of their major shortcomings is the fact that in practice, they tend to produce samples with little diversity, even when trained on diverse datasets. This phenomenon, known as mode collapse, has been the main focus of several recent advances in GANs. Yet there is little understanding of why mode collapse happens and why existing approaches are able to mitigate mode collapse. We propose a principled approach to handling mode collapse, which we call {\\em packing}. The main idea is to modify the discriminator to make decisions based on multiple samples from the same class, either real or artificially generated. We borrow analysis tools from binary hypothesis testing---in particular the seminal result of Blackwell \\cite{Bla53}---to prove a fundamental connection between packing and mode collapse. We show that packing naturally penalizes generators with mode collapse, thereby favoring generator distributions with less mode collapse during the training process. Numerical experiments on benchmark datasets suggests that packing provides significant improvements in practice as well.","Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2018-08-31 without embargo terms","The student, Ashish Kumar Khetan, accepted the attached license on 2018-04-25 at 10:58.","The student, Ashish Kumar Khetan, submitted this Thesis for approval on 2018-04-25 at 11:06.","This Thesis was approved for publication on 2018-04-25 at 13:42.","DSpace SAF Submission Ingestion Package generated from Vireo submission #12456 on 2018-08-31 at 17:14:33","Made available in DSpace on 2018-09-04T20:27:27Z (GMT). No. of bitstreams: 2 KHETAN-THESIS-2018.pdf: 864360 bytes, checksum: 7bd9e0dc7d28b7ce9be1a306630638a9 (MD5) LICENSE.txt: 4216 bytes, checksum: 3a692fbeb8fb8f2ed797c12c22c3e244 (MD5) Previous issue date: 2018-04-25"],"dc:format":["application/pdf"],"dc:identifier":["http://hdl.handle.net/2142/101060"],"dc:language":["en"],"dc:rights":["Copyright 2018 Ashish Kumar Khetan"],"dc:subject":["GAN"],"dc:title":["PacGAN: The power of two samples in generative adversarial networks"],"dc:type":["text"],"thesis:degree_discipline":["Computer Science"],"thesis:degree_level":["Thesis"],"thesis:degree_name":["M.S."],"thesis:institution_name":["University of Illinois at Urbana-Champaign"]},"updated_at":"2026-07-22T22:24:38Z"}