Show simple item record

dc.contributor.authorFrisella, Megan
dc.contributor.authorKhorrami, Pooya
dc.contributor.authorMatterer, Jason
dc.contributor.authorKratkiewicz, Kendra
dc.contributor.authorTorres-Carrasquillo, Pedro
dc.date.accessioned2022-04-25T12:36:10Z
dc.date.available2022-04-25T12:36:10Z
dc.date.issued2022-04-20
dc.identifier.urihttps://hdl.handle.net/1721.1/142034
dc.description.abstractMachine learning models perform face verification (FV) for a variety of highly consequential applications, such as biometric authentication, face identification, and surveillance. Many state-of-the-art FV systems suffer from unequal performance across demographic groups, which is commonly overlooked by evaluation measures that do not assess population-specific performance. Deployed systems with bias may result in serious harm against individuals or groups who experience underperformance. We explore several fairness definitions and metrics, attempting to quantify bias in Google’s FaceNet model. In addition to statistical fairness metrics, we analyze clustered face embeddings produced by the FV model. We link well-clustered embeddings (well-defined, dense clusters) for a demographic group to biased model performance against that group. We present the intuition that FV systems underperform on protected demographic groups because they are less sensitive to differences between features within those groups, as evidenced by clustered embeddings. We show how this performance discrepancy results from a combination of representation and aggregation bias.en_US
dc.publisherMultidisciplinary Digital Publishing Instituteen_US
dc.relation.isversionofhttp://dx.doi.org/10.3390/cmsf2022003006en_US
dc.rightsCreative Commons Attributionen_US
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/en_US
dc.sourceMultidisciplinary Digital Publishing Instituteen_US
dc.titleQuantifying Bias in a Face Verification Systemen_US
dc.typeArticleen_US
dc.identifier.citationComputer Sciences & Mathematics Forum 3 (1): 6 (2022)en_US
dc.contributor.departmentLincoln Laboratory
dc.identifier.mitlicensePUBLISHER_CC
dc.eprint.versionFinal published versionen_US
dc.type.urihttp://purl.org/eprint/type/JournalArticleen_US
eprint.statushttp://purl.org/eprint/status/PeerRevieweden_US
dc.date.updated2022-04-21T21:03:49Z
dspace.date.submission2022-04-21T21:03:49Z
mit.licensePUBLISHER_CC
mit.metadata.statusAuthority Work and Publication Information Neededen_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record