{"id":"https://openalex.org/W7221112964","doi":"https://doi.org/10.48550/arxiv.2610.10666","title":"Explaining the Saliency Map Sparsity of Adversarially-Trained Neural Networks","display_name":"Explaining the Saliency Map Sparsity of Adversarially-Trained Neural Networks","publication_year":2026,"publication_date":"2026-10-07","ids":{"openalex":"https://openalex.org/W7221112964","doi":"https://doi.org/10.48550/arxiv.2610.10666"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.10666","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.10666","pdf_url":"https://arxiv.org/pdf/2610.10666","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.10666","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124984604","display_name":"Yannick Lunk","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I25974101","display_name":"University of W\u00fcrzburg","ror":"https://ror.org/00fbnyb24","country_code":"DE","type":"education","lineage":["https://openalex.org/I25974101"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Lunk, Yannick","raw_affiliation_strings":["Institute of Mathematics University of W\u00fcrzburg , Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Mathematics University of W\u00fcrzburg , Germany","institution_ids":["https://openalex.org/I25974101"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5158213376","display_name":"Atell Yehor Krasnopolsky","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Krasnopolsky, Atell Yehor","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005199568","display_name":"Damien Garreau","orcid":"https://orcid.org/0000-0002-7855-2847","observed_orcids":["https://orcid.org/0000-0002-7855-2847"]},"institutions":[{"id":"https://openalex.org/I25974101","display_name":"University of W\u00fcrzburg","ror":"https://ror.org/00fbnyb24","country_code":"DE","type":"education","lineage":["https://openalex.org/I25974101"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Garreau, Damien","raw_affiliation_strings":["Institute of Computer Science , CAIDAS University of W\u00fcrzburg , Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Institute of Computer Science , CAIDAS University of W\u00fcrzburg , Germany","institution_ids":["https://openalex.org/I25974101"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5035045595","display_name":"Leon Bungert","orcid":"https://orcid.org/0000-0002-6554-9892","observed_orcids":["https://orcid.org/0000-0002-6554-9892"]},"institutions":[{"id":"https://openalex.org/I25974101","display_name":"University of W\u00fcrzburg","ror":"https://ror.org/00fbnyb24","country_code":"DE","type":"education","lineage":["https://openalex.org/I25974101"]},{"id":"https://openalex.org/I62916508","display_name":"Technical University of Munich","ror":"https://ror.org/02kkvpp62","country_code":"DE","type":"education","lineage":["https://openalex.org/I62916508"]}],"countries":["DE"],"is_corresponding":false,"raw_author_name":"Bungert, Leon","raw_affiliation_strings":["Atell Yehor Krasnopolsky Technical University of Munich , Germany","Institute of Mathematics , CAIDAS University of W\u00fcrzburg , Germany"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Atell Yehor Krasnopolsky Technical University of Munich , Germany","institution_ids":["https://openalex.org/I62916508"]},{"raw_affiliation_string":"Institute of Mathematics , CAIDAS University of W\u00fcrzburg , Germany","institution_ids":["https://openalex.org/I25974101"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.5990999937057495,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.5990999937057495,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.38179999589920044,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11612","display_name":"Stochastic Gradient Optimization Techniques","score":0.006200000178068876,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/saliency-maps","display_name":"saliency maps","score":0.9240000247955322},{"id":"https://openalex.org/keywords/gradient-sparsity","display_name":"gradient sparsity","score":0.9120000004768372},{"id":"https://openalex.org/keywords/adversarial-training","display_name":"adversarial training","score":0.9890000224113464},{"id":"https://openalex.org/keywords/explainable-ai","display_name":"explainable AI","score":0.6679999828338623},{"id":"https://openalex.org/keywords/weight-decay","display_name":"weight decay","score":0.6949999928474426},{"id":"https://openalex.org/keywords/bayesian-classifier","display_name":"Bayesian classifier","score":0.8349999785423279},{"id":"https://openalex.org/keywords/empirical-risk-minimization","display_name":"empirical risk minimization","score":0.6190000176429749}],"concepts":[{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6797999739646912},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5806000232696533},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.5712000131607056},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.5591999888420105},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.5432999730110168},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.5315999984741211},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.5231000185012817},{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.5202999711036682},{"id":"https://openalex.org/C2780069185","wikidata":"https://www.wikidata.org/wiki/Q7977945","display_name":"Equivalence (formal languages)","level":2,"score":0.5166000127792358},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.4016000032424927},{"id":"https://openalex.org/C191795146","wikidata":"https://www.wikidata.org/wiki/Q3878446","display_name":"Norm (philosophy)","level":2,"score":0.3971000015735626},{"id":"https://openalex.org/C107321475","wikidata":"https://www.wikidata.org/wiki/Q5374254","display_name":"Empirical risk minimization","level":2,"score":0.392300009727478},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3903999924659729},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.38100001215934753},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.37290000915527344},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3718999922275543},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3610000014305115},{"id":"https://openalex.org/C207201462","wikidata":"https://www.wikidata.org/wiki/Q182505","display_name":"Bayes' theorem","level":3,"score":0.34450000524520874},{"id":"https://openalex.org/C21080849","wikidata":"https://www.wikidata.org/wiki/Q13611879","display_name":"Data point","level":2,"score":0.32260000705718994},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3133000135421753},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.30169999599456787},{"id":"https://openalex.org/C108583219","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep learning","level":2,"score":0.295199990272522},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.2903999984264374},{"id":"https://openalex.org/C135252773","wikidata":"https://www.wikidata.org/wiki/Q1567213","display_name":"Inverse problem","level":2,"score":0.27300000190734863},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.258899986743927},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.2535000145435333}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.10666","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.10666","pdf_url":"https://arxiv.org/pdf/2610.10666","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.10666","doi":"https://doi.org/10.48550/arxiv.2610.10666","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.10666","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.10666","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.10666","pdf_url":"https://arxiv.org/pdf/2610.10666","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G1511613426","display_name":null,"funder_award_id":"544579844","funder_id":"https://openalex.org/F4320320879","funder_display_name":"Deutsche Forschungsgemeinschaft"}],"funders":[{"id":"https://openalex.org/F4320320879","display_name":"Deutsche Forschungsgemeinschaft","ror":"https://ror.org/018mejw64"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7221112964.pdf","grobid_xml":"https://content.openalex.org/works/W7221112964.grobid-xml"},"referenced_works_count":5,"referenced_works":["https://openalex.org/W1988115241","https://openalex.org/W2032849018","https://openalex.org/W2626639386","https://openalex.org/W3116218254","https://openalex.org/W4415799193"],"related_works":["https://openalex.org/W4402727735","https://openalex.org/W2979494945","https://openalex.org/W7134826261","https://openalex.org/W4386556256","https://openalex.org/W2969599988","https://openalex.org/W2944621111","https://openalex.org/W4282829300","https://openalex.org/W3171219085","https://openalex.org/W4410950114","https://openalex.org/W4382763378"],"abstract_inverted_index":{"Understanding":[0],"why":[1],"deep":[2],"neural":[3,51],"networks":[4],"make":[5],"a":[6,30,33,58,126],"given":[7],"prediction":[8],"is":[9,41,146],"of":[10,36,45,49,61,74,80,103,168],"great":[11],"importance":[12],"for":[13,29,64,96,138],"their":[14],"safe":[15],"deployment.":[16],"In":[17,53],"computer":[18],"vision,":[19],"saliency":[20,47],"maps,":[21],"which":[22],"highlight":[23],"the":[24,42,71,78,81,101,110,143,162],"image":[25],"region":[26],"most":[27],"influential":[28],"prediction,":[31],"remain":[32],"widely-used":[34],"form":[35],"explanation.":[37],"An":[38],"empirical":[39,82],"observation":[40],"apparent":[43],"sparsity":[44,167],"gradient":[46,131,144,163],"maps":[48],"adversarially-trained":[50],"networks.":[52,67],"this":[54,62],"paper,":[55],"we":[56,120],"propose":[57],"theoretical":[59,157],"explanation":[60],"phenomenon":[63],"two-layer":[65],"ReLU":[66],"We":[68,154],"build":[69],"on":[70],"established":[72],"equivalence":[73],"adversarial":[75,90,139],"training":[76,140],"to":[77,115,125],"minimization":[79],"risk":[83],"with":[84,129,141],"weight-decay":[85],"penalization":[86],"and":[87,106,109,132,148,165],"an":[88],"added":[89],"total":[91],"variation":[92],"term":[93],"--":[94],"valid":[95],"certain":[97],"loss":[98],"functions.":[99],"As":[100],"number":[102],"data":[104],"points":[105],"neurons":[107],"grows":[108],"regularization":[111],"parameters":[112],"are":[113],"sent":[114],"zero":[116],"at":[117],"appropriate":[118],"rates,":[119],"prove":[121],"that":[122],"minimizers":[123],"converge":[124],"Bayes":[127],"classifier":[128],"minimal":[130],"Barron":[133],"norm.":[134],"Sparsity":[135],"appears":[136],"since":[137],"$\\ell_\\infty$-attacks":[142],"norm":[145],"anisotropic":[147],"favors":[149],"axis-aligned":[150],"/":[151],"sparse":[152],"gradients.":[153],"illustrate":[155],"our":[156],"findings":[158],"experimentally":[159],"by":[160],"evaluating":[161],"$\\ell_1$-norm":[164],"thresholded":[166],"naturally":[169],"versus":[170],"adversarially":[171],"trained":[172],"models.":[173]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-10T00:00:00"}