{"id":"a222994950520d7836c881109b6792cc52d612f841cb0b9cd91b2b1d1807633e","title":"DISCERN authors: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","argument":"The authors present DISCERN, a class unlearning method designed for deletion requests that arrive one after another. In incremental class unlearning, a model must forget each removed class, keep previously forgotten classes forgotten, and preserve the classes that remain. DISCERN replaces access to the retain set—the training data for classes being kept—with synthetic embeddings sampled from per-class parametric distributions. Rather than assume every class follows the same kind of distribution around its centroid, the method tests several Euclidean and directional families and selects among them. The authors say this choice reflects differences in latent geometry between transformer and convolutional networks.\n\nIn experiments on MNIST, CIFAR-10, and CIFAR-100, across four backbones, the authors report that DISCERN was the most stable method in incremental class unlearning on transformers. On ViT-B/16, after 20 sequential deletions, its harmonic mean was 0.797; the strongest method in the simultaneous setting fell from 0.918 to 0.048. A paired ablation, the authors say, found that per-class distribution selection reduced resurfacing of previously forgotten classes. They characterize this as stabilizing accumulated forgetting, not improving every result uniformly. The reported findings concern the datasets, backbones, and unlearning settings evaluated.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports experiments on MNIST, CIFAR-10, and CIFAR-100 with two convolutional and two transformer backbones; it does not establish performance in other settings. The reported results are experimental measurements, not evidence of performance in real deployments.","speaker":{"name":"Segura-Kunsagi, Nicolás, Suárez-Díaz, Juan Luis, Sevillano-García, Iván, Herrera, Francisco","roleAtSource":"Article author"},"source":{"title":"DISCERN: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07892","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors introduce DISCERN, which uses synthetic embeddings sampled from per-class parametric distributions in place of access to the retain set.","excerpt":"replaces access to the retain set with synthetic embeddings sampled from per-class parametric distributions","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"},{"text":"The authors report that DISCERN was the most stable method in incremental class unlearning on transformers in their evaluation.","excerpt":"is the most stable method in ICU on transformers","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"},{"text":"On ViT-B/16 after 20 sequential deletions, the authors report a harmonic mean of 0.797 for DISCERN; the strongest method in the simultaneous setting drops from 0.918 to 0.048.","excerpt":"on ViT-B/16, after 20 sequential deletions it retains a harmonic mean of $0.797$, whereas the strongest method in the simultaneous setting drops from $0.918$ to $0.048$.","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"},{"text":"The authors’ paired ablation suggests that per-class distribution selection helps reduce resurfacing of previously forgotten classes, rather than producing a uniform improvement.","excerpt":"per-class distribution selection substantially reduces the resurfacing of previously forgotten classes, acting as a stabilizer of accumulated forgetting rather than a uniform improvement.","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07892","checkedAt":"2026-11-09T02:02:07.513Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Does DISCERN’s reported reduction in resurfacing of previously forgotten classes hold beyond the datasets, backbones, and unlearning settings evaluated?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"DISCERN authors: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","summary":"The authors present DISCERN, a class unlearning method designed for deletion requests that arrive one after another. In incremental class unlearning, a model must forget each removed class, keep previously forgotten classes forgotten, and preserve the classes that remain. DISCERN replaces access to the retain set—the training data for classes being kept—with synthetic embeddings sampled from per-class parametric distributions. Rather than assume every class follows the same kind of distribution around its centroid, the method tests several Euclidean and directional families and selects among them. The authors say this choice reflects differences in latent geometry between transformer and convolutional networks.\n\nIn experiments on MNIST, CIFAR-10, and CIFAR-100, across four backbones, the authors report that DISCERN was the most stable method in incremental class unlearning on transformers. On ViT-B/16, after 20 sequential deletions, its harmonic mean was 0.797; the strongest method in the simultaneous setting fell from 0.918 to 0.048. A paired ablation, the authors say, found that per-class distribution selection reduced resurfacing of previously forgotten classes. They characterize this as stabilizing accumulated forgetting, not improving every result uniformly. The reported findings concern the datasets, backbones, and unlearning settings evaluated.","question":"Does DISCERN’s reported reduction in resurfacing of previously forgotten classes hold beyond the datasets, backbones, and unlearning settings evaluated?","claims":[{"text":"The authors introduce DISCERN, which uses synthetic embeddings sampled from per-class parametric distributions in place of access to the retain set.","excerpt":"replaces access to the retain set with synthetic embeddings sampled from per-class parametric distributions","support":"source-stated"},{"text":"The authors report that DISCERN was the most stable method in incremental class unlearning on transformers in their evaluation.","excerpt":"is the most stable method in ICU on transformers","support":"source-stated"},{"text":"On ViT-B/16 after 20 sequential deletions, the authors report a harmonic mean of 0.797 for DISCERN; the strongest method in the simultaneous setting drops from 0.918 to 0.048.","excerpt":"on ViT-B/16, after 20 sequential deletions it retains a harmonic mean of $0.797$, whereas the strongest method in the simultaneous setting drops from $0.918$ to $0.048$.","support":"source-stated"},{"text":"The authors’ paired ablation suggests that per-class distribution selection helps reduce resurfacing of previously forgotten classes, rather than producing a uniform improvement.","excerpt":"per-class distribution selection substantially reduces the resurfacing of previously forgotten classes, acting as a stabilizer of accumulated forgetting rather than a uniform improvement.","support":"source-stated"}],"limitations":["The abstract reports experiments on MNIST, CIFAR-10, and CIFAR-100 with two convolutional and two transformer backbones; it does not establish performance in other settings.","The reported results are experimental measurements, not evidence of performance in real deployments."],"mediaId":"fce262c38e421e961689d7608bcb8d611ffb8da82e6c8076823c23ca645830bd","correctionReason":null,"id":"a222994950520d7836c881109b6792cc52d612f841cb0b9cd91b2b1d1807633e","publicationId":"31ba521c-b99b-4f9e-8bfa-3bc55c93c428","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07892","sourceTitle":"DISCERN: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:03:26.791Z","createdAt":"2026-11-09T01:31:23.064Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"fce262c38e421e961689d7608bcb8d611ffb8da82e6c8076823c23ca645830bd","kind":"article","url":"https://arxiv.org/abs/2611.07892","sourceUrl":"https://arxiv.org/abs/2611.07892","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T01:25:15.671Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07892","checkedAt":"2026-11-09T02:02:07.513Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"fce262c38e421e961689d7608bcb8d611ffb8da82e6c8076823c23ca645830bd","confidence":0.79,"provider":"typesafe","model":"jev-1.13.0","inputHash":"e0db4ef88d4fc0da0fcc88b8b45ac2033c4b9eef39d29589a85be2b2e5c4ef88","decidedAt":"2026-11-09T01:25:17.297Z"},"provenance":{"textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observationId":"2fe935f22fe9556af645dfc8166a4a6df12b8737e777fa0d22fabcdac146130b","retrievedAt":"2026-11-09T01:25:15.403Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"b2cf47cf-ee1a-4acc-8c35-cdf3bd95df00"},"reviewer":{"model":"gpt-6-luna","runId":"c3268153-a1c5-4515-8b7b-b66c7620d238"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T01:31:23.064Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Segura-Kunsagi, Nicolás","url":null,"kind":"unknown"},{"name":"Suárez-Díaz, Juan Luis","url":null,"kind":"unknown"},{"name":"Sevillano-García, Iván","url":null,"kind":"unknown"},{"name":"Herrera, Francisco","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Segura-Kunsagi, Nicolás","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Segura-Kunsagi, Nicolás","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}},{"actor":{"name":"Suárez-Díaz, Juan Luis","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Suárez-Díaz, Juan Luis","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}},{"actor":{"name":"Sevillano-García, Iván","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sevillano-García, Iván","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}},{"actor":{"name":"Herrera, Francisco","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Herrera, Francisco","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07892","snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z","basis":"publisher-page-metadata","sourceLinks":44,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/a222994950520d7836c881109b6792cc52d612f841cb0b9cd91b2b1d1807633e/source"}},"canonicalUrl":"https://escapemeridian.com/readings/discern-authors-distributional-synthetic-replay-for-retain-free-a2229949","formats":{"html":"/readings/discern-authors-distributional-synthetic-replay-for-retain-free-a2229949","markdown":"/readings/discern-authors-distributional-synthetic-replay-for-retain-free-a2229949.md","json":"/readings/discern-authors-distributional-synthetic-replay-for-retain-free-a2229949.json"},"attribution":"Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.","maintainedGuideLinks":[],"readerContext":{"version":"reader-context-v1","attribution":"Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.","terms":[{"id":"term-embedding-vector","name":"Embedding vector","explanation":"A list of numbers that represents a piece of text, an image or another item so software can compare meanings.","referenceUrl":"https://developers.google.com/machine-learning/glossary#embedding-vector"}],"entities":[{"id":"arxiv","name":"arXiv","kind":"organization","context":"arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.","referenceUrls":["https://arxiv.org/","https://info.arxiv.org/help/api/index.html"],"path":"/entities/arxiv","json":"/entities/arxiv.json","markdown":"/entities/arxiv.md"},{"id":"harmonic","name":"Harmonic","kind":"organization","context":"Harmonic develops systems that generate mathematical reasoning and proofs, including Aristotle.","referenceUrls":["https://www.harmonic.fun/","https://arxiv.org/abs/2510.01346v2"],"path":"/entities/harmonic","json":"/entities/harmonic.json","markdown":"/entities/harmonic.md"}]},"researchQuestions":[],"withdrawn":false,"references":{"schemaVersion":1,"target":{"type":"reading","id":"a222994950520d7836c881109b6792cc52d612f841cb0b9cd91b2b1d1807633e","revision":"31ba521c-b99b-4f9e-8bfa-3bc55c93c428","revisionKind":"publication-id","title":"DISCERN authors: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","url":"https://escapemeridian.com/readings/discern-authors-distributional-synthetic-replay-for-retain-free-a2229949"},"source":{"type":"source","id":"https://arxiv.org/abs/2611.07892","revision":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","revisionKind":"sha256-source-text","title":"DISCERN: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","url":"https://arxiv.org/abs/2611.07892"},"status":"available","items":[],"next":null,"scope":"Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter."},"revisions":[{"id":"a222994950520d7836c881109b6792cc52d612f841cb0b9cd91b2b1d1807633e","title":"DISCERN authors: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","argument":"The authors present DISCERN, a class unlearning method designed for deletion requests that arrive one after another. In incremental class unlearning, a model must forget each removed class, keep previously forgotten classes forgotten, and preserve the classes that remain. DISCERN replaces access to the retain set—the training data for classes being kept—with synthetic embeddings sampled from per-class parametric distributions. Rather than assume every class follows the same kind of distribution around its centroid, the method tests several Euclidean and directional families and selects among them. The authors say this choice reflects differences in latent geometry between transformer and convolutional networks.\n\nIn experiments on MNIST, CIFAR-10, and CIFAR-100, across four backbones, the authors report that DISCERN was the most stable method in incremental class unlearning on transformers. On ViT-B/16, after 20 sequential deletions, its harmonic mean was 0.797; the strongest method in the simultaneous setting fell from 0.918 to 0.048. A paired ablation, the authors say, found that per-class distribution selection reduced resurfacing of previously forgotten classes. They characterize this as stabilizing accumulated forgetting, not improving every result uniformly. The reported findings concern the datasets, backbones, and unlearning settings evaluated.","topicIds":["intelligence-and-discovery"],"domain":"science","claimType":"attributed reading","limits":"The abstract reports experiments on MNIST, CIFAR-10, and CIFAR-100 with two convolutional and two transformer backbones; it does not establish performance in other settings. The reported results are experimental measurements, not evidence of performance in real deployments.","speaker":{"name":"Segura-Kunsagi, Nicolás, Suárez-Díaz, Juan Luis, Sevillano-García, Iván, Herrera, Francisco","roleAtSource":"Article author"},"source":{"title":"DISCERN: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","publisher":"arXiv cs.AI","format":"Article","url":"https://arxiv.org/abs/2611.07892","publishedAt":"2026-11-05","dateBasis":"Date basis not provided."},"evidence":[{"text":"The authors introduce DISCERN, which uses synthetic embeddings sampled from per-class parametric distributions in place of access to the retain set.","excerpt":"replaces access to the retain set with synthetic embeddings sampled from per-class parametric distributions","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"},{"text":"The authors report that DISCERN was the most stable method in incremental class unlearning on transformers in their evaluation.","excerpt":"is the most stable method in ICU on transformers","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"},{"text":"On ViT-B/16 after 20 sequential deletions, the authors report a harmonic mean of 0.797 for DISCERN; the strongest method in the simultaneous setting drops from 0.918 to 0.048.","excerpt":"on ViT-B/16, after 20 sequential deletions it retains a harmonic mean of $0.797$, whereas the strongest method in the simultaneous setting drops from $0.918$ to $0.048$.","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"},{"text":"The authors’ paired ablation suggests that per-class distribution selection helps reduce resurfacing of previously forgotten classes, rather than producing a uniform improvement.","excerpt":"per-class distribution selection substantially reduces the resurfacing of previously forgotten classes, acting as a stabilizer of accumulated forgetting rather than a uniform improvement.","url":"https://arxiv.org/abs/2611.07892","relationship":"Source-stated"}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07892","checkedAt":"2026-11-09T02:02:07.513Z","items":[],"note":"No suitable media was found by the supported page extraction."},"assumptions":[],"openQuestion":{"attribution":"Open question","text":"Does DISCERN’s reported reduction in resurfacing of previously forgotten classes hold beyond the datasets, backbones, and unlearning settings evaluated?"},"whatWouldChangeView":{"attribution":"Change condition not recorded","text":""},"editorialComparisons":[],"live":{"title":"DISCERN authors: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","summary":"The authors present DISCERN, a class unlearning method designed for deletion requests that arrive one after another. In incremental class unlearning, a model must forget each removed class, keep previously forgotten classes forgotten, and preserve the classes that remain. DISCERN replaces access to the retain set—the training data for classes being kept—with synthetic embeddings sampled from per-class parametric distributions. Rather than assume every class follows the same kind of distribution around its centroid, the method tests several Euclidean and directional families and selects among them. The authors say this choice reflects differences in latent geometry between transformer and convolutional networks.\n\nIn experiments on MNIST, CIFAR-10, and CIFAR-100, across four backbones, the authors report that DISCERN was the most stable method in incremental class unlearning on transformers. On ViT-B/16, after 20 sequential deletions, its harmonic mean was 0.797; the strongest method in the simultaneous setting fell from 0.918 to 0.048. A paired ablation, the authors say, found that per-class distribution selection reduced resurfacing of previously forgotten classes. They characterize this as stabilizing accumulated forgetting, not improving every result uniformly. The reported findings concern the datasets, backbones, and unlearning settings evaluated.","question":"Does DISCERN’s reported reduction in resurfacing of previously forgotten classes hold beyond the datasets, backbones, and unlearning settings evaluated?","claims":[{"text":"The authors introduce DISCERN, which uses synthetic embeddings sampled from per-class parametric distributions in place of access to the retain set.","excerpt":"replaces access to the retain set with synthetic embeddings sampled from per-class parametric distributions","support":"source-stated"},{"text":"The authors report that DISCERN was the most stable method in incremental class unlearning on transformers in their evaluation.","excerpt":"is the most stable method in ICU on transformers","support":"source-stated"},{"text":"On ViT-B/16 after 20 sequential deletions, the authors report a harmonic mean of 0.797 for DISCERN; the strongest method in the simultaneous setting drops from 0.918 to 0.048.","excerpt":"on ViT-B/16, after 20 sequential deletions it retains a harmonic mean of $0.797$, whereas the strongest method in the simultaneous setting drops from $0.918$ to $0.048$.","support":"source-stated"},{"text":"The authors’ paired ablation suggests that per-class distribution selection helps reduce resurfacing of previously forgotten classes, rather than producing a uniform improvement.","excerpt":"per-class distribution selection substantially reduces the resurfacing of previously forgotten classes, acting as a stabilizer of accumulated forgetting rather than a uniform improvement.","support":"source-stated"}],"limitations":["The abstract reports experiments on MNIST, CIFAR-10, and CIFAR-100 with two convolutional and two transformer backbones; it does not establish performance in other settings.","The reported results are experimental measurements, not evidence of performance in real deployments."],"mediaId":"fce262c38e421e961689d7608bcb8d611ffb8da82e6c8076823c23ca645830bd","correctionReason":null,"id":"a222994950520d7836c881109b6792cc52d612f841cb0b9cd91b2b1d1807633e","publicationId":"31ba521c-b99b-4f9e-8bfa-3bc55c93c428","version":1,"sourceUrl":"https://arxiv.org/abs/2611.07892","sourceTitle":"DISCERN: Distributional Synthetic Replay for Retain-Free Incremental Machine Unlearning","publisher":"arXiv cs.AI","publishedAt":null,"originalPublished":{"value":"2026-11-05","precision":"day","basis":"publisher_citation_date"},"firstObservedAt":"2026-11-06T05:03:26.791Z","createdAt":"2026-11-09T01:31:23.064Z","historical":true,"entityIds":["arxiv"],"topicIds":["intelligence-and-discovery"],"media":[{"id":"fce262c38e421e961689d7608bcb8d611ffb8da82e6c8076823c23ca645830bd","kind":"article","url":"https://arxiv.org/abs/2611.07892","sourceUrl":"https://arxiv.org/abs/2611.07892","label":"Read original","provenance":"source_link","verification":{"status":"verified","checkedAt":"2026-11-09T01:25:15.671Z","method":"acquired_source","contentType":"text/html"},"recommendationReason":"Recommended by Jev for understanding this source; checked by Luna."}],"sourceMedia":{"status":"not_found","sourceUrl":"https://arxiv.org/abs/2611.07892","checkedAt":"2026-11-09T02:02:07.513Z","items":[],"note":"No suitable media was found by the supported page extraction."},"editorialDecision":{"outcome":"accept","reasonCodes":["material_development"],"rubricVersion":"stream-admission-v5","mediaId":"fce262c38e421e961689d7608bcb8d611ffb8da82e6c8076823c23ca645830bd","confidence":0.79,"provider":"typesafe","model":"jev-1.13.0","inputHash":"e0db4ef88d4fc0da0fcc88b8b45ac2033c4b9eef39d29589a85be2b2e5c4ef88","decidedAt":"2026-11-09T01:25:17.297Z"},"provenance":{"textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observationId":"2fe935f22fe9556af645dfc8166a4a6df12b8737e777fa0d22fabcdac146130b","retrievedAt":"2026-11-09T01:25:15.403Z","kind":"publisher-page","preparer":{"model":"gpt-6-luna","runId":"b2cf47cf-ee1a-4acc-8c35-cdf3bd95df00"},"reviewer":{"model":"gpt-6-luna","runId":"c3268153-a1c5-4515-8b7b-b66c7620d238"}},"audit":{"status":"uncertain","reviewedAt":"2026-11-09T01:31:23.064Z","model":"anthropic/claude-opus-5.5"},"sourceAttribution":{"authors":[{"name":"Segura-Kunsagi, Nicolás","url":null,"kind":"unknown"},{"name":"Suárez-Díaz, Juan Luis","url":null,"kind":"unknown"},{"name":"Sevillano-García, Iván","url":null,"kind":"unknown"},{"name":"Herrera, Francisco","url":null,"kind":"unknown"}],"authorRole":"article-author","hosts":[],"status":"stated","authorRoster":{"truncated":false,"omitted":0,"assertionGroupsOmitted":0},"participation":[{"actor":{"name":"Segura-Kunsagi, Nicolás","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Segura-Kunsagi, Nicolás","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:0","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}},{"actor":{"name":"Suárez-Díaz, Juan Luis","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Suárez-Díaz, Juan Luis","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:1","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}},{"actor":{"name":"Sevillano-García, Iván","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Sevillano-García, Iván","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:2","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}},{"actor":{"name":"Herrera, Francisco","url":null,"kind":"unknown","actorId":null},"role":"author","scope":{"kind":"work-credit","locator":null},"literalCredit":"Herrera, Francisco","status":"stated","provenance":{"basis":"head-meta","sourceUrl":"https://arxiv.org/abs/2611.07892","locator":"head/meta[name=author|citation_author]"},"assertionId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0:participation:3","identity":{"version":"entity-identity-v1","status":"unresolved","actorId":null,"canonicalActorId":null,"candidateActorIds":[],"assertions":[],"sourceObservedAt":"2026-11-09T02:02:07.513Z","knowledgeAsOf":null},"sourceVersion":{"snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","sourceUrl":"https://arxiv.org/abs/2611.07892","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z"}}],"participationCoverage":{"total":4,"returned":4,"complete":true,"fullAssertions":"source-metadata-endpoint"},"distribution":null,"publisher":null,"sourceUrl":"https://arxiv.org/abs/2611.07892","snapshotId":"5d7d532e954235fa52908baa4581a82ab289693be70ad771888a98783c0dc3d0","contentSha256":"57f0f806ee017ac6086e2a0bde364ad662ddcbbd77432282ddcccd3a674f7566","textHash":"2af3e1d224e0988fc7d560da0c2cc06a0829ba2627cbeab49239c1dbb9a274a8","observedAt":"2026-11-09T02:02:07.513Z","basis":"publisher-page-metadata","sourceLinks":44,"linksTruncated":false,"metadataUrl":"/api/pipeline/stream/a222994950520d7836c881109b6792cc52d612f841cb0b9cd91b2b1d1807633e/source"}}}],"audits":[{"publicationId":"31ba521c-b99b-4f9e-8bfa-3bc55c93c428","verdict":"uncertain","findings":[{"dimension":"attribution","status":"pass","explanation":"The reading attributes DISCERN and all reported results to the paper's authors, matching the arXiv listing. Results are framed as what the authors report or say, not as established fact.","excerpt":"Authors:Nicolás Segura-Kunsagi, Juan Luis Suárez-Díaz, Iván Sevillano-García, Francisco Herrera"},{"dimension":"summary","status":"uncertain","explanation":"The summary is faithful, but a non-specialist cannot readily follow the central idea or result. It never says plainly what machine unlearning is: removing a training data's influence without retraining from scratch. Key terms go unexplained in plain language, including synthetic embeddings, parametric distributions, Euclidean/directional families, latent geometry, backbones and ablation. Above all, it does not say what the harmonic mean measures, so the headline figures 0.797 and 0.048 are hard to interpret. Smallest repair: add a one-clause definition of unlearning and of what the harmonic mean balances, or ensure term-library tooltips cover these.","excerpt":"Machine unlearning aims to remove the influence of specific training data from a trained model without retraining from scratch."},{"dimension":"date","status":"pass","explanation":"The originalPublished value of 2026-11-05 matches the v1 submission date on the arXiv page. The later feed timestamp is correctly not used.","excerpt":"[Submitted on 5 Nov 2026]"},{"dimension":"support","status":"pass","explanation":"Each claim matches the abstract: retain-set replacement, stability in ICU on transformers, the ViT-B/16 figures, and the ablation characterization. The limitations scope results to the tested datasets and backbones without overreach. The question is a genuine open issue and not a leading claim. No unsupported benefits or costs are implied.","excerpt":"DISCERN is competitive when classes are removed at once and is the most stable method in ICU on transformers"},{"dimension":"media","status":"pass","explanation":"The selected media is the verified direct arXiv abstract page for this paper. That is the most direct access to the source.","excerpt":null}],"createdAt":"2026-11-09T01:31:23.064Z","model":"anthropic/claude-opus-5.5"}],"correctionUrl":"https://escapemeridian.com/corrections/"}