PARC: A Quantitative Framework Uncovering the Symmetries within Vision Language Models Article Swipe

PDF

Jenny Schmalfuß , Nadine Chang , Vibashan VS , Maying Shen , Andrés Bruhn , Jose M. Álvarez ·

YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2506.14808

Vision language models (VLMs) respond to user-crafted text prompts and visual inputs, and are applied to numerous real-world problems. VLMs integrate visual modalities with large language models (LLMs), which are well known to be prompt-sensitive. Hence, it is crucial to determine whether VLMs inherit this instability to varying prompts. We therefore investigate which prompt variations VLMs are most sensitive to and which VLMs are most agnostic to prompt variations. To this end, we introduce PARC (Prompt Analysis via Reliability and Calibration), a VLM prompt sensitivity analysis framework built on three pillars: (1) plausible prompt variations in both the language and vision domain, (2) a novel model reliability score with built-in guarantees, and (3) a calibration step that enables dataset- and prompt-spanning prompt variation analysis. Regarding prompt variations, PARC's evaluation shows that VLMs mirror LLM language prompt sensitivity in the vision domain, and most destructive variations change the expected answer. Regarding models, outstandingly robust VLMs among 22 evaluated models come from the InternVL2 family. We further find indications that prompt sensitivity is linked to training data. The code will be at https://github.com/NVlabs/PARC.

Related Topics

National Institutional Ranking Framework

The Myth Of The Framework

Cynefin Framework

Parc Des Princes

Quantitative Easing

Django (Web Framework)

Iran Nuclear Deal Framework

United Nations Framework Convention On Climate Change

Bootstrap (Front-End Framework)

Parc Olympique Lyonnais

Quantitative Research

Electron (Software Framework)

Concepts

No concepts available.

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2506.14808
PDF: https://arxiv.org/pdf/2506.14808
OA Status: green
OpenAlex ID: https://openalex.org/W4415332906

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4415332906

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2506.14808

Digital Object Identifier
Title: PARC: A Quantitative Framework Uncovering the Symmetries within Vision Language Models

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2025

Year of publication
Publication date: 2025-06-03

Full publication date if available
Authors: Jenny Schmalfuß, Nadine Chang, Vibashan VS, Maying Shen, Andrés Bruhn, Jose M. Álvarez

List of authors in order
Landing page: https://arxiv.org/abs/2506.14808

Publisher landing page
PDF URL: https://arxiv.org/pdf/2506.14808

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/2506.14808

Direct OA link when available
Cited by: 0

Total citation count in OpenAlex

Full payload

id	https://openalex.org/W4415332906
doi	https://doi.org/10.48550/arxiv.2506.14808
ids.doi	https://doi.org/10.48550/arxiv.2506.14808
ids.openalex	https://openalex.org/W4415332906
fwci
type	preprint
title	PARC: A Quantitative Framework Uncovering the Symmetries within Vision Language Models
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
topics[0].id	https://openalex.org/T12694
topics[0].field.id	https://openalex.org/fields/32
topics[0].field.display_name	Psychology
topics[0].score	0.8174999952316284
topics[0].domain.id	https://openalex.org/domains/2
topics[0].domain.display_name	Social Sciences
topics[0].subfield.id	https://openalex.org/subfields/3205
topics[0].subfield.display_name	Experimental and Cognitive Psychology
topics[0].display_name	Categorization, perception, and language
is_xpac	False
apc_list
apc_paid
language	en
locations[0].id	pmh:oai:arXiv.org:2506.14808
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/2506.14808
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/2506.14808
locations[1].id	doi:10.48550/arxiv.2506.14808
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.2506.14808
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5014373655
authorships[0].author.orcid	https://orcid.org/0000-0001-8507-927X
authorships[0].author.display_name	Jenny Schmalfuß
authorships[0].author_position	first
authorships[0].raw_author_name	Schmalfuss, Jenny
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5045190437
authorships[1].author.orcid	https://orcid.org/0000-0003-4765-8478
authorships[1].author.display_name	Nadine Chang
authorships[1].author_position	middle
authorships[1].raw_author_name	Chang, Nadine
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5004330920
authorships[2].author.orcid
authorships[2].author.display_name	Vibashan VS
authorships[2].author_position	middle
authorships[2].raw_author_name	VS, Vibashan
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5079403120
authorships[3].author.orcid
authorships[3].author.display_name	Maying Shen
authorships[3].author_position	middle
authorships[3].raw_author_name	Shen, Maying
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5018641843
authorships[4].author.orcid	https://orcid.org/0000-0003-0423-7411
authorships[4].author.display_name	Andrés Bruhn
authorships[4].author_position	middle
authorships[4].raw_author_name	Bruhn, Andres
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5101540588
authorships[5].author.orcid	https://orcid.org/0000-0002-7535-6322
authorships[5].author.display_name	Jose M. Álvarez
authorships[5].author_position	last
authorships[5].raw_author_name	Alvarez, Jose M.
authorships[5].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/2506.14808
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-19T00:00:00
display_name	PARC: A Quantitative Framework Uncovering the Symmetries within Vision Language Models
has_fulltext	False
is_retracted	False
updated_date	2025-11-06T06:51:31.235846
primary_topic.id	https://openalex.org/T12694
primary_topic.field.id	https://openalex.org/fields/32
primary_topic.field.display_name	Psychology
primary_topic.score	0.8174999952316284
primary_topic.domain.id	https://openalex.org/domains/2
primary_topic.domain.display_name	Social Sciences
primary_topic.subfield.id	https://openalex.org/subfields/3205
primary_topic.subfield.display_name	Experimental and Cognitive Psychology
primary_topic.display_name	Categorization, perception, and language
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2506.14808
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/2506.14808
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/2506.14808
primary_location.id	pmh:oai:arXiv.org:2506.14808
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/2506.14808
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/2506.14808
publication_date	2025-06-03
publication_year	2025
referenced_works_count	0
abstract_inverted_index.a	81, 103, 113
abstract_inverted_index.22	155
abstract_inverted_index.To	69
abstract_inverted_index.We	49, 163
abstract_inverted_index.at	179
abstract_inverted_index.be	33, 178
abstract_inverted_index.in	95, 137
abstract_inverted_index.is	37, 170
abstract_inverted_index.it	36
abstract_inverted_index.on	88
abstract_inverted_index.to	5, 15, 32, 39, 46, 59, 66, 172
abstract_inverted_index.we	72
abstract_inverted_index.(1)	91
abstract_inverted_index.(2)	102
abstract_inverted_index.(3)	112
abstract_inverted_index.LLM	133
abstract_inverted_index.The	175
abstract_inverted_index.VLM	82
abstract_inverted_index.and	9, 12, 60, 79, 99, 111, 119, 141
abstract_inverted_index.are	13, 29, 56, 63
abstract_inverted_index.the	97, 138, 146, 160
abstract_inverted_index.via	77
abstract_inverted_index.PARC	74
abstract_inverted_index.VLMs	19, 42, 55, 62, 131, 153
abstract_inverted_index.both	96
abstract_inverted_index.code	176
abstract_inverted_index.come	158
abstract_inverted_index.end,	71
abstract_inverted_index.find	165
abstract_inverted_index.from	159
abstract_inverted_index.most	57, 64, 142
abstract_inverted_index.step	115
abstract_inverted_index.text	7
abstract_inverted_index.that	116, 130, 167
abstract_inverted_index.this	44, 70
abstract_inverted_index.well	30
abstract_inverted_index.will	177
abstract_inverted_index.with	23, 108
abstract_inverted_index.among	154
abstract_inverted_index.built	87
abstract_inverted_index.data.	174
abstract_inverted_index.known	31
abstract_inverted_index.large	24
abstract_inverted_index.model	105
abstract_inverted_index.novel	104
abstract_inverted_index.score	107
abstract_inverted_index.shows	129
abstract_inverted_index.three	89
abstract_inverted_index.which	28, 52, 61
abstract_inverted_index.(VLMs)	3
abstract_inverted_index.Hence,	35
abstract_inverted_index.PARC's	127
abstract_inverted_index.Vision	0
abstract_inverted_index.change	145
abstract_inverted_index.linked	171
abstract_inverted_index.mirror	132
abstract_inverted_index.models	2, 26, 157
abstract_inverted_index.prompt	53, 67, 83, 93, 121, 125, 135, 168
abstract_inverted_index.robust	152
abstract_inverted_index.vision	100, 139
abstract_inverted_index.visual	10, 21
abstract_inverted_index.(LLMs),	27
abstract_inverted_index.(Prompt	75
abstract_inverted_index.answer.	148
abstract_inverted_index.applied	14
abstract_inverted_index.crucial	38
abstract_inverted_index.domain,	101, 140
abstract_inverted_index.enables	117
abstract_inverted_index.family.	162
abstract_inverted_index.further	164
abstract_inverted_index.inherit	43
abstract_inverted_index.inputs,	11
abstract_inverted_index.models,	150
abstract_inverted_index.prompts	8
abstract_inverted_index.respond	4
abstract_inverted_index.varying	47
abstract_inverted_index.whether	41
abstract_inverted_index.Analysis	76
abstract_inverted_index.agnostic	65
abstract_inverted_index.analysis	85
abstract_inverted_index.built-in	109
abstract_inverted_index.dataset-	118
abstract_inverted_index.expected	147
abstract_inverted_index.language	1, 25, 98, 134
abstract_inverted_index.numerous	16
abstract_inverted_index.pillars:	90
abstract_inverted_index.prompts.	48
abstract_inverted_index.training	173
abstract_inverted_index.InternVL2	161
abstract_inverted_index.Regarding	124, 149
abstract_inverted_index.analysis.	123
abstract_inverted_index.determine	40
abstract_inverted_index.evaluated	156
abstract_inverted_index.framework	86
abstract_inverted_index.integrate	20
abstract_inverted_index.introduce	73
abstract_inverted_index.plausible	92
abstract_inverted_index.problems.	18
abstract_inverted_index.sensitive	58
abstract_inverted_index.therefore	50
abstract_inverted_index.variation	122
abstract_inverted_index.evaluation	128
abstract_inverted_index.modalities	22
abstract_inverted_index.real-world	17
abstract_inverted_index.variations	54, 94, 144
abstract_inverted_index.Reliability	78
abstract_inverted_index.calibration	114
abstract_inverted_index.destructive	143
abstract_inverted_index.guarantees,	110
abstract_inverted_index.indications	166
abstract_inverted_index.instability	45
abstract_inverted_index.investigate	51
abstract_inverted_index.reliability	106
abstract_inverted_index.sensitivity	84, 136, 169
abstract_inverted_index.variations,	126
abstract_inverted_index.variations.	68
abstract_inverted_index.user-crafted	6
abstract_inverted_index.Calibration),	80
abstract_inverted_index.outstandingly	151
abstract_inverted_index.prompt-spanning	120
abstract_inverted_index.prompt-sensitive.	34
abstract_inverted_index.https://github.com/NVlabs/PARC.	180
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	6
citation_normalized_percentile