{"id":{"repo_id":"mit","oai_identifier":"oai:dspace.mit.edu:1721.1/130717"},"canonical_url":"https://search.dev.ndltd.org/etd/mit/oai:dspace.mit.edu:1721.1/130717","repository":{"repo_id":"mit","name":"MIT","base_url":"https://dspace.mit.edu/oai/request"},"display":{"title":"Contrasting contrastive and supervised models interpretability","abstract":"In this thesis, we compare the representations of an unsupervised contrastive model to those of an equivalent supervised model using several deep neural network interpretability methods: network dissection, sparsity experiments, and saliency maps. Network dissections of self-supervised contrastive and supervised models show that the neurons of the contrastive model tend to learn about different parts of an object (ie. top-half of a dog or left-half of a person) while the neurons of the supervised model tend to learn about the entire object (ie. a dog or a person). Sparsity experiments show that the representations learned by the contrastive model are less sparse than the representations learned by the supervised counterpart model. Saliency maps show that the contrastive model focuses more on specific parts of the input image. Finally, we find that the contrastive model representations transfer better to finegrained classification tasks than the supervised model representations.","abstract_html":"In this thesis, we compare the representations of an unsupervised contrastive model to those of an equivalent supervised model using several deep neural network interpretability methods: network dissection, sparsity experiments, and saliency maps. Network dissections of self-supervised contrastive and supervised models show that the neurons of the contrastive model tend to learn about different parts of an object (ie. top-half of a dog or left-half of a person) while the neurons of the supervised model tend to learn about the entire object (ie. a dog or a person). Sparsity experiments show that the representations learned by the contrastive model are less sparse than the representations learned by the supervised counterpart model. Saliency maps show that the contrastive model focuses more on specific parts of the input image. Finally, we find that the contrastive model representations transfer better to finegrained classification tasks than the supervised model representations.","abstract_has_math":false,"creators":["You, Yejin."],"institution":"Massachusetts Institute of Technology","degree_name":"Master","degree_level":null,"degree_discipline":null,"degree_department":"Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science","school":null,"contributors":[],"advisors":["Antonio Torralba and David Bau."],"committee_chairs":[],"committee_members":[],"year":2021,"date_issued":"2021","date_published":"2021","updated_at":"2026-07-22T22:20:59Z","subjects":["Electrical Engineering and Computer Science."],"languages":["eng"],"rights":["MIT theses may be protected by copyright. Please reuse MIT thesis content according to the MIT Libraries Permissions Policy, which is available through the URL provided."],"rights_urls":["http://dspace.mit.edu/handle/1721.1/7582"],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/1721.1/130717","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Antonio Torralba and David Bau."]},{"key":"dc:contributor.department","label":"Department","values":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science","EECS"]},{"key":"dc:contributor.other","label":"Dc Contributor Other","values":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science."]},{"key":"dc:creator","label":"Author","values":["You, Yejin."]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2021-05-24T19:52:49Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2021-05-24T19:52:49Z"]},{"key":"dc:date.issued","label":"Date","values":["2021"]},{"key":"dc:publisher","label":"Institution","values":["Massachusetts Institute of Technology"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Electrical Engineering and Computer Science."]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["MIT theses may be protected by copyright. Please reuse MIT thesis content according to the MIT Libraries Permissions Policy, which is available through the URL provided."]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://dspace.mit.edu/handle/1721.1/7582"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/1721.1/130717"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, February, 2021","Cataloged from the official PDF of thesis.","Includes bibliographical references (pages 61-62)."]},{"key":"dc:description.abstract","label":"Abstract","values":["In this thesis, we compare the representations of an unsupervised contrastive model to those of an equivalent supervised model using several deep neural network interpretability methods: network dissection, sparsity experiments, and saliency maps. Network dissections of self-supervised contrastive and supervised models show that the neurons of the contrastive model tend to learn about different parts of an object (ie. top-half of a dog or left-half of a person) while the neurons of the supervised model tend to learn about the entire object (ie. a dog or a person). Sparsity experiments show that the representations learned by the contrastive model are less sparse than the representations learned by the supervised counterpart model. Saliency maps show that the contrastive model focuses more on specific parts of the input image. Finally, we find that the contrastive model representations transfer better to finegrained classification tasks than the supervised model representations."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["M. Eng."]},{"key":"dc:title","label":"Title","values":["Contrasting contrastive and supervised models interpretability"]}]}],"canonical_facts":{"dc:contributor.advisor":["Antonio Torralba and David Bau."],"dc:contributor.department":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science","EECS"],"dc:contributor.other":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science."],"dc:creator":["You, Yejin."],"dc:date.accessioned":["2021-05-24T19:52:49Z"],"dc:date.available":["2021-05-24T19:52:49Z"],"dc:date.issued":["2021"],"dc:description":["Thesis: M. Eng., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, February, 2021","Cataloged from the official PDF of thesis.","Includes bibliographical references (pages 61-62)."],"dc:description.abstract":["In this thesis, we compare the representations of an unsupervised contrastive model to those of an equivalent supervised model using several deep neural network interpretability methods: network dissection, sparsity experiments, and saliency maps. Network dissections of self-supervised contrastive and supervised models show that the neurons of the contrastive model tend to learn about different parts of an object (ie. top-half of a dog or left-half of a person) while the neurons of the supervised model tend to learn about the entire object (ie. a dog or a person). Sparsity experiments show that the representations learned by the contrastive model are less sparse than the representations learned by the supervised counterpart model. Saliency maps show that the contrastive model focuses more on specific parts of the input image. Finally, we find that the contrastive model representations transfer better to finegrained classification tasks than the supervised model representations."],"dc:description.degree":["M. Eng."],"dc:identifier.uri":["https://hdl.handle.net/1721.1/130717"],"dc:language.iso":["eng"],"dc:publisher":["Massachusetts Institute of Technology"],"dc:rights":["MIT theses may be protected by copyright. Please reuse MIT thesis content according to the MIT Libraries Permissions Policy, which is available through the URL provided."],"dc:rights.uri":["http://dspace.mit.edu/handle/1721.1/7582"],"dc:subject":["Electrical Engineering and Computer Science."],"dc:title":["Contrasting contrastive and supervised models interpretability"],"dc:type":["Thesis"],"thesis:degree_name":["Master"]},"updated_at":"2026-07-22T22:20:59Z"}