Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation Article Swipe

PDF

Anh Tuan Bui , Long Vuong , Khanh Doan , Trung Le , Paul Montague , Tamas Abraham , Dinh Phung ·

YOU? · · 2024 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2410.15618

Diffusion models excel at generating visually striking content from text but can inadvertently produce undesirable or harmful content when trained on unfiltered internet data. A practical solution is to selectively removing target concepts from the model, but this may impact the remaining concepts. Prior approaches have tried to balance this by introducing a loss term to preserve neutral content or a regularization term to minimize changes in the model parameters, yet resolving this trade-off remains challenging. In this work, we propose to identify and preserving concepts most affected by parameter changes, termed as \textit{adversarial concepts}. This approach ensures stable erasure with minimal impact on the other concepts. We demonstrate the effectiveness of our method using the Stable Diffusion model, showing that it outperforms state-of-the-art erasure methods in eliminating unwanted content while maintaining the integrity of other unrelated elements. Our code is available at https://github.com/tuananhbui89/Erasing-Adversarial-Preservation.

Related Topics

Computer Science

Artificial Intelligence

Physics

Concepts

Adversarial system Diffusion Computer science Artificial intelligence Physics Thermodynamics

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2410.15618
PDF: https://arxiv.org/pdf/2410.15618
OA Status: green
Related Works: 10
OpenAlex ID: https://openalex.org/W4404342051

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4404342051

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2410.15618

Digital Object Identifier
Title: Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2024

Year of publication
Publication date: 2024-10-21

Full publication date if available
Authors: Anh Tuan Bui, Long Vuong, Khanh Doan, Trung Le, Paul Montague, Tamas Abraham, Dinh Phung

List of authors in order
Landing page: https://arxiv.org/abs/2410.15618

Publisher landing page
PDF URL: https://arxiv.org/pdf/2410.15618

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2410.15618

Direct OA link when available
Concepts: Adversarial system, Diffusion, Computer science, Artificial intelligence, Physics, Thermodynamics

Top concepts (fields/topics) attached by OpenAlex
Cited by: 0

Total citation count in OpenAlex
Related works (count): 10

Other works algorithmically related by OpenAlex

Full payload

id	https://openalex.org/W4404342051
doi	https://doi.org/10.48550/arxiv.2410.15618
ids.doi	https://doi.org/10.48550/arxiv.2410.15618
ids.openalex	https://openalex.org/W4404342051
fwci
type	preprint
title	Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
topics[0].id	https://openalex.org/T11689
topics[0].field.id	https://openalex.org/fields/17
topics[0].field.display_name	Computer Science
topics[0].score	0.9901999831199646
topics[0].domain.id	https://openalex.org/domains/3
topics[0].domain.display_name	Physical Sciences
topics[0].subfield.id	https://openalex.org/subfields/1702
topics[0].subfield.display_name	Artificial Intelligence
topics[0].display_name	Adversarial Robustness in Machine Learning
is_xpac	False
apc_list
apc_paid
concepts[0].id	https://openalex.org/C37736160
concepts[0].level	2
concepts[0].score	0.7934786081314087
concepts[0].wikidata	https://www.wikidata.org/wiki/Q1801315
concepts[0].display_name	Adversarial system
concepts[1].id	https://openalex.org/C69357855
concepts[1].level	2
concepts[1].score	0.5892102718353271
concepts[1].wikidata	https://www.wikidata.org/wiki/Q163214
concepts[1].display_name	Diffusion
concepts[2].id	https://openalex.org/C41008148
concepts[2].level	0
concepts[2].score	0.5147145390510559
concepts[2].wikidata	https://www.wikidata.org/wiki/Q21198
concepts[2].display_name	Computer science
concepts[3].id	https://openalex.org/C154945302
concepts[3].level	1
concepts[3].score	0.27431437373161316
concepts[3].wikidata	https://www.wikidata.org/wiki/Q11660
concepts[3].display_name	Artificial intelligence
concepts[4].id	https://openalex.org/C121332964
concepts[4].level	0
concepts[4].score	0.08296170830726624
concepts[4].wikidata	https://www.wikidata.org/wiki/Q413
concepts[4].display_name	Physics
concepts[5].id	https://openalex.org/C97355855
concepts[5].level	1
concepts[5].score	0.06354838609695435
concepts[5].wikidata	https://www.wikidata.org/wiki/Q11473
concepts[5].display_name	Thermodynamics
keywords[0].id	https://openalex.org/keywords/adversarial-system
keywords[0].score	0.7934786081314087
keywords[0].display_name	Adversarial system
keywords[1].id	https://openalex.org/keywords/diffusion
keywords[1].score	0.5892102718353271
keywords[1].display_name	Diffusion
keywords[2].id	https://openalex.org/keywords/computer-science
keywords[2].score	0.5147145390510559
keywords[2].display_name	Computer science
keywords[3].id	https://openalex.org/keywords/artificial-intelligence
keywords[3].score	0.27431437373161316
keywords[3].display_name	Artificial intelligence
keywords[4].id	https://openalex.org/keywords/physics
keywords[4].score	0.08296170830726624
keywords[4].display_name	Physics
keywords[5].id	https://openalex.org/keywords/thermodynamics
keywords[5].score	0.06354838609695435
keywords[5].display_name	Thermodynamics
language	en
locations[0].id	pmh:oai:arXiv.org:2410.15618
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2410.15618
locations[0].version	submittedVersion
locations[0].raw_type	text
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2410.15618
locations[1].id	doi:10.48550/arxiv.2410.15618
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license	cc-by
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id	https://openalex.org/licenses/cc-by
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2410.15618
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5032421586
authorships[0].author.orcid	https://orcid.org/0000-0002-4153-5074
authorships[0].author.display_name	Anh Tuan Bui
authorships[0].author_position	first
authorships[0].raw_author_name	Bui, Anh
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5002307757
authorships[1].author.orcid
authorships[1].author.display_name	Long Vuong
authorships[1].author_position	middle
authorships[1].raw_author_name	Vuong, Long
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5053288613
authorships[2].author.orcid
authorships[2].author.display_name	Khanh Doan
authorships[2].author_position	middle
authorships[2].raw_author_name	Doan, Khanh
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5046456210
authorships[3].author.orcid	https://orcid.org/0009-0008-5254-0160
authorships[3].author.display_name	Trung Le
authorships[3].author_position	middle
authorships[3].raw_author_name	Le, Trung
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5091338477
authorships[4].author.orcid
authorships[4].author.display_name	Paul Montague
authorships[4].author_position	middle
authorships[4].raw_author_name	Montague, Paul
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5077639290
authorships[5].author.orcid	https://orcid.org/0000-0003-2466-7646
authorships[5].author.display_name	Tamas Abraham
authorships[5].author_position	middle
authorships[5].raw_author_name	Abraham, Tamas
authorships[5].is_corresponding	False
authorships[6].author.id	https://openalex.org/A5036447132
authorships[6].author.orcid	https://orcid.org/0000-0002-9977-8247
authorships[6].author.display_name	Dinh Phung
authorships[6].author_position	last
authorships[6].raw_author_name	Phung, Dinh
authorships[6].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2410.15618
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-10T00:00:00
display_name	Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation
has_fulltext	False
is_retracted	False
updated_date	2025-11-06T06:51:31.235846
primary_topic.id	https://openalex.org/T11689
primary_topic.field.id	https://openalex.org/fields/17
primary_topic.field.display_name	Computer Science
primary_topic.score	0.9901999831199646
primary_topic.domain.id	https://openalex.org/domains/3
primary_topic.domain.display_name	Physical Sciences
primary_topic.subfield.id	https://openalex.org/subfields/1702
primary_topic.subfield.display_name	Artificial Intelligence
primary_topic.display_name	Adversarial Robustness in Machine Learning
related_works	https://openalex.org/W4391375266, https://openalex.org/W2899084033, https://openalex.org/W2748952813, https://openalex.org/W2502115930, https://openalex.org/W2482350142, https://openalex.org/W4246396837, https://openalex.org/W3126451824, https://openalex.org/W1561927205, https://openalex.org/W3191453585, https://openalex.org/W4297672492
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2410.15618
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2410.15618
best_oa_location.version	submittedVersion
best_oa_location.raw_type	text
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2410.15618
primary_location.id	pmh:oai:arXiv.org:2410.15618
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2410.15618
primary_location.version	submittedVersion
primary_location.raw_type	text
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2410.15618
publication_date	2024-10-21
publication_year	2024
referenced_works_count	0
abstract_inverted_index.A	24
abstract_inverted_index.a	52, 60
abstract_inverted_index.In	76
abstract_inverted_index.We	107
abstract_inverted_index.as	92
abstract_inverted_index.at	3, 142
abstract_inverted_index.by	50, 88
abstract_inverted_index.in	66, 126
abstract_inverted_index.is	27, 140
abstract_inverted_index.it	121
abstract_inverted_index.of	111, 134
abstract_inverted_index.on	20, 103
abstract_inverted_index.or	15, 59
abstract_inverted_index.to	28, 47, 55, 63, 81
abstract_inverted_index.we	79
abstract_inverted_index.Our	138
abstract_inverted_index.and	83
abstract_inverted_index.but	10, 36
abstract_inverted_index.can	11
abstract_inverted_index.may	38
abstract_inverted_index.our	112
abstract_inverted_index.the	34, 40, 67, 104, 109, 115, 132
abstract_inverted_index.yet	70
abstract_inverted_index.This	95
abstract_inverted_index.code	139
abstract_inverted_index.from	8, 33
abstract_inverted_index.have	45
abstract_inverted_index.loss	53
abstract_inverted_index.most	86
abstract_inverted_index.term	54, 62
abstract_inverted_index.text	9
abstract_inverted_index.that	120
abstract_inverted_index.this	37, 49, 72, 77
abstract_inverted_index.when	18
abstract_inverted_index.with	100
abstract_inverted_index.Prior	43
abstract_inverted_index.data.	23
abstract_inverted_index.excel	2
abstract_inverted_index.model	68
abstract_inverted_index.other	105, 135
abstract_inverted_index.tried	46
abstract_inverted_index.using	114
abstract_inverted_index.while	130
abstract_inverted_index.work,	78
abstract_inverted_index.Stable	116
abstract_inverted_index.impact	39, 102
abstract_inverted_index.method	113
abstract_inverted_index.model,	35, 118
abstract_inverted_index.models	1
abstract_inverted_index.stable	98
abstract_inverted_index.target	31
abstract_inverted_index.termed	91
abstract_inverted_index.balance	48
abstract_inverted_index.changes	65
abstract_inverted_index.content	7, 17, 58, 129
abstract_inverted_index.ensures	97
abstract_inverted_index.erasure	99, 124
abstract_inverted_index.harmful	16
abstract_inverted_index.methods	125
abstract_inverted_index.minimal	101
abstract_inverted_index.neutral	57
abstract_inverted_index.produce	13
abstract_inverted_index.propose	80
abstract_inverted_index.remains	74
abstract_inverted_index.showing	119
abstract_inverted_index.trained	19
abstract_inverted_index.affected	87
abstract_inverted_index.approach	96
abstract_inverted_index.changes,	90
abstract_inverted_index.concepts	32, 85
abstract_inverted_index.identify	82
abstract_inverted_index.internet	22
abstract_inverted_index.minimize	64
abstract_inverted_index.preserve	56
abstract_inverted_index.removing	30
abstract_inverted_index.solution	26
abstract_inverted_index.striking	6
abstract_inverted_index.unwanted	128
abstract_inverted_index.visually	5
abstract_inverted_index.Diffusion	0, 117
abstract_inverted_index.available	141
abstract_inverted_index.concepts.	42, 106
abstract_inverted_index.elements.	137
abstract_inverted_index.integrity	133
abstract_inverted_index.parameter	89
abstract_inverted_index.practical	25
abstract_inverted_index.remaining	41
abstract_inverted_index.resolving	71
abstract_inverted_index.trade-off	73
abstract_inverted_index.unrelated	136
abstract_inverted_index.approaches	44
abstract_inverted_index.concepts}.	94
abstract_inverted_index.generating	4
abstract_inverted_index.preserving	84
abstract_inverted_index.unfiltered	21
abstract_inverted_index.demonstrate	108
abstract_inverted_index.eliminating	127
abstract_inverted_index.introducing	51
abstract_inverted_index.maintaining	131
abstract_inverted_index.outperforms	122
abstract_inverted_index.parameters,	69
abstract_inverted_index.selectively	29
abstract_inverted_index.undesirable	14
abstract_inverted_index.challenging.	75
abstract_inverted_index.effectiveness	110
abstract_inverted_index.inadvertently	12
abstract_inverted_index.regularization	61
abstract_inverted_index.state-of-the-art	123
abstract_inverted_index.\textit{adversarial	93
abstract_inverted_index.https://github.com/tuananhbui89/Erasing-Adversarial-Preservation.	143
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	7
citation_normalized_percentile