{"id":"https://openalex.org/W7155041074","doi":"https://doi.org/10.48550/arxiv.2604.16358","title":"SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamics","display_name":"SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamics","publication_year":2026,"publication_date":"2026-03-18","ids":{"openalex":"https://openalex.org/W7155041074","doi":"https://doi.org/10.48550/arxiv.2604.16358"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2604.16358","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.16358","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2604.16358","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5134098425","display_name":"Haolong Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Haolong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134146033","display_name":"Hanyu Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Hanyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134122378","display_name":"Tiancheng He","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Tiancheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060429045","display_name":"Huahui Yi","orcid":"https://orcid.org/0009-0007-9361-5491"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yi, Huahui","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134123831","display_name":"An Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, An","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134188672","display_name":"Qiankun Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Qiankun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134204315","display_name":"Kun Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Kun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134197918","display_name":"Yang Liu","orcid":"https://orcid.org/0009-0007-3811-6163"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5134197349","display_name":"Zhigang Zeng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zeng, Zhigang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"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":0.4779999852180481,"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":0.4779999852180481,"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/T11241","display_name":"Advanced Malware Detection Techniques","score":0.12520000338554382,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T13295","display_name":"Safety Systems Engineering in Autonomy","score":0.047600001096725464,"subfield":{"id":"https://openalex.org/subfields/2213","display_name":"Safety, Risk, Reliability and Quality"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/exploit","display_name":"Exploit","score":0.8090000152587891},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.6959999799728394},{"id":"https://openalex.org/keywords/bootstrapping","display_name":"Bootstrapping (finance)","score":0.6067000031471252},{"id":"https://openalex.org/keywords/trajectory","display_name":"Trajectory","score":0.4560999870300293},{"id":"https://openalex.org/keywords/scaling","display_name":"Scaling","score":0.3993000090122223},{"id":"https://openalex.org/keywords/rare-events","display_name":"Rare events","score":0.38440001010894775},{"id":"https://openalex.org/keywords/adversarial-system","display_name":"Adversarial system","score":0.3662000000476837},{"id":"https://openalex.org/keywords/synthetic-data","display_name":"Synthetic data","score":0.3488999903202057}],"concepts":[{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.8090000152587891},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7597000002861023},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.6959999799728394},{"id":"https://openalex.org/C207609745","wikidata":"https://www.wikidata.org/wiki/Q4944086","display_name":"Bootstrapping (finance)","level":2,"score":0.6067000031471252},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.4560999870300293},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4357999861240387},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4212999939918518},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.41819998621940613},{"id":"https://openalex.org/C99844830","wikidata":"https://www.wikidata.org/wiki/Q102441924","display_name":"Scaling","level":2,"score":0.3993000090122223},{"id":"https://openalex.org/C2777317252","wikidata":"https://www.wikidata.org/wiki/Q18393516","display_name":"Rare events","level":2,"score":0.38440001010894775},{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.3662000000476837},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.3488999903202057},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3391999900341034},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.33640000224113464},{"id":"https://openalex.org/C2776544517","wikidata":"https://www.wikidata.org/wiki/Q189447","display_name":"Unexpected events","level":2,"score":0.33309999108314514},{"id":"https://openalex.org/C2779662365","wikidata":"https://www.wikidata.org/wiki/Q5416694","display_name":"Event (particle physics)","level":2,"score":0.30320000648498535},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.30219998955726624},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.28619998693466187},{"id":"https://openalex.org/C133462117","wikidata":"https://www.wikidata.org/wiki/Q4929239","display_name":"Data collection","level":2,"score":0.27230000495910645},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.2639999985694885},{"id":"https://openalex.org/C145912823","wikidata":"https://www.wikidata.org/wiki/Q113558","display_name":"Dynamics (music)","level":2,"score":0.2549999952316284},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.2517000138759613}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2604.16358","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.16358","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2604.16358","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2604.16358","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"MLLMs":[0],"are":[1],"increasingly":[2],"deployed":[3],"in":[4],"multi-turn":[5,51,103,142],"settings,":[6],"where":[7],"attackers":[8],"can":[9],"escalate":[10],"unsafe":[11],"intent":[12],"through":[13],"the":[14,80,94],"evolving":[15],"visual-text":[16],"history":[17],"and":[18,32,40,83,112,141,159],"exploit":[19],"long-context":[20],"safety":[21,24,105],"decay.":[22],"Yet":[23],"alignment":[25,52],"is":[26,166],"still":[27],"dominated":[28],"by":[29],"single-turn":[30,130],"data":[31],"fixed-template":[33],"dialogues,":[34],"leaving":[35],"a":[36,49,64,102],"mismatch":[37],"between":[38],"training":[39],"deployment.":[41],"To":[42],"bridge":[43],"this":[44],"gap,":[45],"we":[46],"propose":[47],"SaFeR-Steer,":[48],"progressive":[50],"framework":[53],"that":[54,89],"combines":[55],"staged":[56],"synthetic":[57],"bootstrapping":[58],"with":[59,107],"tutor-in-the-loop":[60],"GRPO":[61],"to":[62,156],"train":[63],"single":[65],"student":[66],"under":[67],"adaptive,":[68],"on-policy":[69],"attacks.":[70],"We":[71,99],"also":[72],"introduce":[73],"Trajectory-Consistent":[74],"Summative":[75],"Reward":[76],"(TCSR),":[77],"which":[78],"aggregates":[79],"historical":[81],"minimum":[82],"average":[84],"of":[85],"turn":[86,92],"rewards":[87],"so":[88],"any":[90],"low-quality":[91],"affects":[93],"trajectory-level":[95],"return.":[96],"I.":[97],"Dataset.":[98],"release":[100],"STEER,":[101],"multimodal":[104],"dataset":[106],"STEER-SFT":[108],"(12,934),":[109],"STEER-RL":[110],"(2,000),":[111],"STEER-Bench":[113],"(3,227)":[114],"dialogues":[115],"spanning":[116],"2-10":[117],"turns.":[118],"II.":[119],"Experiment.":[120],"Starting":[121],"from":[122],"Qwen2.5-VL-3B/7B,":[123],"SaFeR-Steer":[124],"substantially":[125],"improves":[126],"Safety/Helpfulness":[127],"on":[128],"both":[129],"(48.30/45.86":[131],"$\\rightarrow$":[132,137,145,150],"81.84/70.77":[133],"for":[134,139,147,152],"3B;":[135,148],"56.21/60.32":[136],"87.89/77.40":[138],"7B)":[140],"benchmarks":[143],"(12.55/27.13":[144],"55.58/70.27":[146],"24.66/46.48":[149],"64.89/72.35":[151],"7B),":[153],"shifting":[154],"failures":[155],"later":[157],"turns":[158],"yielding":[160],"robustness":[161],"beyond":[162],"scaling":[163],"alone.":[164],"Code":[165],"available":[167],"at":[168],"https://anonymous.4open.science/r/SaFeR-Steer":[169]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-04-22T00:00:00"}