Towards Robust Evaluation of Unlearning in LLMs via Data Transformations Article Swipe

PDF

Abhinav Joshi , Sujit Saha , D. K. Shukla , Sriram Vema , Harsh Jhamtani , Manas Gaur , Ashutosh Modi ·

YOU? · · 2024 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2411.15477

Large Language Models (LLMs) have shown to be a great success in a wide range of applications ranging from regular NLP-based use cases to AI agents. LLMs have been trained on a vast corpus of texts from various sources; despite the best efforts during the data pre-processing stage while training the LLMs, they may pick some undesirable information such as personally identifiable information (PII). Consequently, in recent times research in the area of Machine Unlearning (MUL) has become active, the main idea is to force LLMs to forget (unlearn) certain information (e.g., PII) without suffering from performance loss on regular tasks. In this work, we examine the robustness of the existing MUL techniques for their ability to enable leakage-proof forgetting in LLMs. In particular, we examine the effect of data transformation on forgetting, i.e., is an unlearned LLM able to recall forgotten information if there is a change in the format of the input? Our findings on the TOFU dataset highlight the necessity of using diverse data formats to quantify unlearning in LLMs more reliably.

Related Topics

Computer Science

Concepts

Political science Computer science

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2411.15477
PDF: https://arxiv.org/pdf/2411.15477
OA Status: green
Related Works: 10
OpenAlex ID: https://openalex.org/W4404986573

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4404986573

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2411.15477

Digital Object Identifier
Title: Towards Robust Evaluation of Unlearning in LLMs via Data Transformations

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2024

Year of publication
Publication date: 2024-11-23

Full publication date if available
Authors: Abhinav Joshi, Sujit Saha, D. K. Shukla, Sriram Vema, Harsh Jhamtani, Manas Gaur, Ashutosh Modi

List of authors in order
Landing page: https://arxiv.org/abs/2411.15477

Publisher landing page
PDF URL: https://arxiv.org/pdf/2411.15477

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/2411.15477

Direct OA link when available
Concepts: Political science, Computer science

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/W4404986573
doi	https://doi.org/10.48550/arxiv.2411.15477
ids.doi	https://doi.org/10.48550/arxiv.2411.15477
ids.openalex	https://openalex.org/W4404986573
fwci
type	preprint
title	Towards Robust Evaluation of Unlearning in LLMs via Data Transformations
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
topics[0].id	https://openalex.org/T10876
topics[0].field.id	https://openalex.org/fields/22
topics[0].field.display_name	Engineering
topics[0].score	0.7224000096321106
topics[0].domain.id	https://openalex.org/domains/3
topics[0].domain.display_name	Physical Sciences
topics[0].subfield.id	https://openalex.org/subfields/2207
topics[0].subfield.display_name	Control and Systems Engineering
topics[0].display_name	Fault Detection and Control Systems
topics[1].id	https://openalex.org/T11801
topics[1].field.id	https://openalex.org/fields/22
topics[1].field.display_name	Engineering
topics[1].score	0.6578999757766724
topics[1].domain.id	https://openalex.org/domains/3
topics[1].domain.display_name	Physical Sciences
topics[1].subfield.id	https://openalex.org/subfields/2212
topics[1].subfield.display_name	Ocean Engineering
topics[1].display_name	Reservoir Engineering and Simulation Methods
topics[2].id	https://openalex.org/T14470
topics[2].field.id	https://openalex.org/fields/22
topics[2].field.display_name	Engineering
topics[2].score	0.6001999974250793
topics[2].domain.id	https://openalex.org/domains/3
topics[2].domain.display_name	Physical Sciences
topics[2].subfield.id	https://openalex.org/subfields/2207
topics[2].subfield.display_name	Control and Systems Engineering
topics[2].display_name	Advanced Data Processing Techniques
is_xpac	False
apc_list
apc_paid
concepts[0].id	https://openalex.org/C17744445
concepts[0].level	0
concepts[0].score	0.33818620443344116
concepts[0].wikidata	https://www.wikidata.org/wiki/Q36442
concepts[0].display_name	Political science
concepts[1].id	https://openalex.org/C41008148
concepts[1].level	0
concepts[1].score	0.328423410654068
concepts[1].wikidata	https://www.wikidata.org/wiki/Q21198
concepts[1].display_name	Computer science
keywords[0].id	https://openalex.org/keywords/political-science
keywords[0].score	0.33818620443344116
keywords[0].display_name	Political science
keywords[1].id	https://openalex.org/keywords/computer-science
keywords[1].score	0.328423410654068
keywords[1].display_name	Computer science
language	en
locations[0].id	pmh:oai:arXiv.org:2411.15477
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/2411.15477
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/2411.15477
locations[1].id	doi:10.48550/arxiv.2411.15477
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
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id
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.2411.15477
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5114847525
authorships[0].author.orcid
authorships[0].author.display_name	Abhinav Joshi
authorships[0].author_position	first
authorships[0].raw_author_name	Joshi, Abhinav
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5008388681
authorships[1].author.orcid	https://orcid.org/0000-0002-9705-5045
authorships[1].author.display_name	Sujit Saha
authorships[1].author_position	middle
authorships[1].raw_author_name	Saha, Shaswati
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5031140914
authorships[2].author.orcid	https://orcid.org/0000-0001-8835-9269
authorships[2].author.display_name	D. K. Shukla
authorships[2].author_position	middle
authorships[2].raw_author_name	Shukla, Divyaksh
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5114847526
authorships[3].author.orcid
authorships[3].author.display_name	Sriram Vema
authorships[3].author_position	middle
authorships[3].raw_author_name	Vema, Sriram
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5114847527
authorships[4].author.orcid
authorships[4].author.display_name	Harsh Jhamtani
authorships[4].author_position	middle
authorships[4].raw_author_name	Jhamtani, Harsh
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5023667301
authorships[5].author.orcid	https://orcid.org/0000-0002-5411-2230
authorships[5].author.display_name	Manas Gaur
authorships[5].author_position	middle
authorships[5].raw_author_name	Gaur, Manas
authorships[5].is_corresponding	False
authorships[6].author.id	https://openalex.org/A5076043215
authorships[6].author.orcid	https://orcid.org/0000-0002-0962-8350
authorships[6].author.display_name	Ashutosh Modi
authorships[6].author_position	last
authorships[6].raw_author_name	Modi, Ashutosh
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/2411.15477
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-10T00:00:00
display_name	Towards Robust Evaluation of Unlearning in LLMs via Data Transformations
has_fulltext	False
is_retracted	False
updated_date	2025-11-06T06:51:31.235846
primary_topic.id	https://openalex.org/T10876
primary_topic.field.id	https://openalex.org/fields/22
primary_topic.field.display_name	Engineering
primary_topic.score	0.7224000096321106
primary_topic.domain.id	https://openalex.org/domains/3
primary_topic.domain.display_name	Physical Sciences
primary_topic.subfield.id	https://openalex.org/subfields/2207
primary_topic.subfield.display_name	Control and Systems Engineering
primary_topic.display_name	Fault Detection and Control Systems
related_works	https://openalex.org/W4391375266, https://openalex.org/W2899084033, https://openalex.org/W2748952813, https://openalex.org/W2390279801, https://openalex.org/W4391913857, https://openalex.org/W2358668433, https://openalex.org/W4396701345, https://openalex.org/W2376932109, https://openalex.org/W2001405890, https://openalex.org/W4396696052
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2411.15477
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/2411.15477
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/2411.15477
primary_location.id	pmh:oai:arXiv.org:2411.15477
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/2411.15477
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/2411.15477
publication_date	2024-11-23
publication_year	2024
referenced_works_count	0
abstract_inverted_index.a	8, 12, 31, 146
abstract_inverted_index.AI	24
abstract_inverted_index.In	101, 122
abstract_inverted_index.an	135
abstract_inverted_index.as	59
abstract_inverted_index.be	7
abstract_inverted_index.if	143
abstract_inverted_index.in	11, 65, 69, 120, 148, 171
abstract_inverted_index.is	82, 134, 145
abstract_inverted_index.of	15, 34, 72, 108, 128, 151, 163
abstract_inverted_index.on	30, 98, 131, 156
abstract_inverted_index.to	6, 23, 83, 86, 116, 139, 168
abstract_inverted_index.we	104, 124
abstract_inverted_index.LLM	137
abstract_inverted_index.MUL	111
abstract_inverted_index.Our	154
abstract_inverted_index.for	113
abstract_inverted_index.has	76
abstract_inverted_index.may	53
abstract_inverted_index.the	40, 44, 50, 70, 79, 106, 109, 126, 149, 152, 157, 161
abstract_inverted_index.use	21
abstract_inverted_index.LLMs	26, 85, 172
abstract_inverted_index.PII)	92
abstract_inverted_index.TOFU	158
abstract_inverted_index.able	138
abstract_inverted_index.area	71
abstract_inverted_index.been	28
abstract_inverted_index.best	41
abstract_inverted_index.data	45, 129, 166
abstract_inverted_index.from	18, 36, 95
abstract_inverted_index.have	4, 27
abstract_inverted_index.idea	81
abstract_inverted_index.loss	97
abstract_inverted_index.main	80
abstract_inverted_index.more	173
abstract_inverted_index.pick	54
abstract_inverted_index.some	55
abstract_inverted_index.such	58
abstract_inverted_index.they	52
abstract_inverted_index.this	102
abstract_inverted_index.vast	32
abstract_inverted_index.wide	13
abstract_inverted_index.(MUL)	75
abstract_inverted_index.LLMs,	51
abstract_inverted_index.LLMs.	121
abstract_inverted_index.Large	0
abstract_inverted_index.cases	22
abstract_inverted_index.force	84
abstract_inverted_index.great	9
abstract_inverted_index.i.e.,	133
abstract_inverted_index.range	14
abstract_inverted_index.shown	5
abstract_inverted_index.stage	47
abstract_inverted_index.texts	35
abstract_inverted_index.their	114
abstract_inverted_index.there	144
abstract_inverted_index.times	67
abstract_inverted_index.using	164
abstract_inverted_index.while	48
abstract_inverted_index.work,	103
abstract_inverted_index.(LLMs)	3
abstract_inverted_index.(PII).	63
abstract_inverted_index.(e.g.,	91
abstract_inverted_index.Models	2
abstract_inverted_index.become	77
abstract_inverted_index.change	147
abstract_inverted_index.corpus	33
abstract_inverted_index.during	43
abstract_inverted_index.effect	127
abstract_inverted_index.enable	117
abstract_inverted_index.forget	87
abstract_inverted_index.format	150
abstract_inverted_index.input?	153
abstract_inverted_index.recall	140
abstract_inverted_index.recent	66
abstract_inverted_index.tasks.	100
abstract_inverted_index.Machine	73
abstract_inverted_index.ability	115
abstract_inverted_index.active,	78
abstract_inverted_index.agents.	25
abstract_inverted_index.certain	89
abstract_inverted_index.dataset	159
abstract_inverted_index.despite	39
abstract_inverted_index.diverse	165
abstract_inverted_index.efforts	42
abstract_inverted_index.examine	105, 125
abstract_inverted_index.formats	167
abstract_inverted_index.ranging	17
abstract_inverted_index.regular	19, 99
abstract_inverted_index.success	10
abstract_inverted_index.trained	29
abstract_inverted_index.various	37
abstract_inverted_index.without	93
abstract_inverted_index.Language	1
abstract_inverted_index.existing	110
abstract_inverted_index.findings	155
abstract_inverted_index.quantify	169
abstract_inverted_index.research	68
abstract_inverted_index.sources;	38
abstract_inverted_index.training	49
abstract_inverted_index.(unlearn)	88
abstract_inverted_index.NLP-based	20
abstract_inverted_index.forgotten	141
abstract_inverted_index.highlight	160
abstract_inverted_index.necessity	162
abstract_inverted_index.reliably.	174
abstract_inverted_index.suffering	94
abstract_inverted_index.unlearned	136
abstract_inverted_index.Unlearning	74
abstract_inverted_index.forgetting	119
abstract_inverted_index.personally	60
abstract_inverted_index.robustness	107
abstract_inverted_index.techniques	112
abstract_inverted_index.unlearning	170
abstract_inverted_index.forgetting,	132
abstract_inverted_index.information	57, 62, 90, 142
abstract_inverted_index.particular,	123
abstract_inverted_index.performance	96
abstract_inverted_index.undesirable	56
abstract_inverted_index.applications	16
abstract_inverted_index.identifiable	61
abstract_inverted_index.Consequently,	64
abstract_inverted_index.leakage-proof	118
abstract_inverted_index.pre-processing	46
abstract_inverted_index.transformation	130
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	7
citation_normalized_percentile