Normalized training loss vs. baselines (Facebook).
<p>DCOR reports both reconstruction-only and total (with RLC); baselines report reconstruction-only. Each curve is normalized as in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0335135#pone.0335135.e243" target="_blank">Eq (29)</a> by di...
Saved in:
| Hovedforfatter: | |
|---|---|
| Andre forfattere: | , , |
| Udgivet: |
2025
|
| Fag: | |
| Tags: |
Tilføj Tag
Ingen Tags, Vær først til at tagge denne postø!
|
| _version_ | 1849927640795840512 |
|---|---|
| author | Hossein Rafieizadeh (22676722) |
| author2 | Hadi Zare (20073000) Mohsen Ghassemi Parsa (22676725) Hocine Cherifi (8177628) |
| author2_role | author author author |
| author_facet | Hossein Rafieizadeh (22676722) Hadi Zare (20073000) Mohsen Ghassemi Parsa (22676725) Hocine Cherifi (8177628) |
| author_role | author |
| dc.creator.none.fl_str_mv | Hossein Rafieizadeh (22676722) Hadi Zare (20073000) Mohsen Ghassemi Parsa (22676725) Hocine Cherifi (8177628) |
| dc.date.none.fl_str_mv | 2025-11-24T18:37:57Z |
| dc.identifier.none.fl_str_mv | 10.1371/journal.pone.0335135.g005 |
| dc.relation.none.fl_str_mv | https://figshare.com/articles/figure/Normalized_training_loss_vs_baselines_Facebook_/30698014 |
| dc.rights.none.fl_str_mv | CC BY 4.0 info:eu-repo/semantics/openAccess |
| dc.subject.none.fl_str_mv | Cell Biology Science Policy Environmental Sciences not elsewhere classified Biological Sciences not elsewhere classified Information Systems not elsewhere classified intrusions across social reconstructions across views level contrastive learning dual contrastive learning across six benchmarks div >< p view discrepancies underutilized augmented graph views dcor improves auroc view discrepancies level contrast dual autoencoder augmented view specific information six datasets reduces auroc publicly available preserves fine physical domains performing non maximum gain leaving cross identifying threats financial fraud existing graph dcor reconstructs dcor ), contrasts reconstructions attributed networks attribute patterns |
| dc.title.none.fl_str_mv | Normalized training loss vs. baselines (Facebook). |
| dc.type.none.fl_str_mv | Image Figure info:eu-repo/semantics/publishedVersion image |
| description | <p>DCOR reports both reconstruction-only and total (with RLC); baselines report reconstruction-only. Each curve is normalized as in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0335135#pone.0335135.e243" target="_blank">Eq (29)</a> by dividing by its epoch-1 value and EMA-smoothed (exponential moving average) with , where the EMA is computed as with . This normalization enables fair visual comparison across methods with different objectives and scales; the plot therefore emphasizes relative convergence trends (shape and stability) rather than raw magnitudes. Consistent with DCOR’s design, RLC regularizes late-phase training: the reconstruction curve decreases more conservatively than methods that minimize reconstruction alone, while the total objective continues to decrease.</p> |
| eu_rights_str_mv | openAccess |
| id | Manara_d47601a13dea14d921a3ef3acb5e6aa2 |
| identifier_str_mv | 10.1371/journal.pone.0335135.g005 |
| network_acronym_str | Manara |
| network_name_str | ManaraRepo |
| oai_identifier_str | oai:figshare.com:article/30698014 |
| publishDate | 2025 |
| repository.mail.fl_str_mv | |
| repository.name.fl_str_mv | |
| repository_id_str | |
| rights_invalid_str_mv | CC BY 4.0 |
| spelling | Normalized training loss vs. baselines (Facebook).Hossein Rafieizadeh (22676722)Hadi Zare (20073000)Mohsen Ghassemi Parsa (22676725)Hocine Cherifi (8177628)Cell BiologyScience PolicyEnvironmental Sciences not elsewhere classifiedBiological Sciences not elsewhere classifiedInformation Systems not elsewhere classifiedintrusions across socialreconstructions across viewslevel contrastive learningdual contrastive learningacross six benchmarksdiv >< pview discrepancies underutilizedaugmented graph viewsdcor improves aurocview discrepancieslevel contrastdual autoencoderaugmented viewspecific informationsix datasetsreduces aurocpublicly availablepreserves finephysical domainsperforming nonmaximum gainleaving crossidentifying threatsfinancial fraudexisting graphdcor reconstructsdcor ),contrasts reconstructionsattributed networksattribute patterns<p>DCOR reports both reconstruction-only and total (with RLC); baselines report reconstruction-only. Each curve is normalized as in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0335135#pone.0335135.e243" target="_blank">Eq (29)</a> by dividing by its epoch-1 value and EMA-smoothed (exponential moving average) with , where the EMA is computed as with . This normalization enables fair visual comparison across methods with different objectives and scales; the plot therefore emphasizes relative convergence trends (shape and stability) rather than raw magnitudes. Consistent with DCOR’s design, RLC regularizes late-phase training: the reconstruction curve decreases more conservatively than methods that minimize reconstruction alone, while the total objective continues to decrease.</p>2025-11-24T18:37:57ZImageFigureinfo:eu-repo/semantics/publishedVersionimage10.1371/journal.pone.0335135.g005https://figshare.com/articles/figure/Normalized_training_loss_vs_baselines_Facebook_/30698014CC BY 4.0info:eu-repo/semantics/openAccessoai:figshare.com:article/306980142025-11-24T18:37:57Z |
| spellingShingle | Normalized training loss vs. baselines (Facebook). Hossein Rafieizadeh (22676722) Cell Biology Science Policy Environmental Sciences not elsewhere classified Biological Sciences not elsewhere classified Information Systems not elsewhere classified intrusions across social reconstructions across views level contrastive learning dual contrastive learning across six benchmarks div >< p view discrepancies underutilized augmented graph views dcor improves auroc view discrepancies level contrast dual autoencoder augmented view specific information six datasets reduces auroc publicly available preserves fine physical domains performing non maximum gain leaving cross identifying threats financial fraud existing graph dcor reconstructs dcor ), contrasts reconstructions attributed networks attribute patterns |
| status_str | publishedVersion |
| title | Normalized training loss vs. baselines (Facebook). |
| title_full | Normalized training loss vs. baselines (Facebook). |
| title_fullStr | Normalized training loss vs. baselines (Facebook). |
| title_full_unstemmed | Normalized training loss vs. baselines (Facebook). |
| title_short | Normalized training loss vs. baselines (Facebook). |
| title_sort | Normalized training loss vs. baselines (Facebook). |
| topic | Cell Biology Science Policy Environmental Sciences not elsewhere classified Biological Sciences not elsewhere classified Information Systems not elsewhere classified intrusions across social reconstructions across views level contrastive learning dual contrastive learning across six benchmarks div >< p view discrepancies underutilized augmented graph views dcor improves auroc view discrepancies level contrast dual autoencoder augmented view specific information six datasets reduces auroc publicly available preserves fine physical domains performing non maximum gain leaving cross identifying threats financial fraud existing graph dcor reconstructs dcor ), contrasts reconstructions attributed networks attribute patterns |