{"id":"https://openalex.org/W7148635577","doi":"https://doi.org/10.5281/zenodo.19398708","title":"The Rotation-Retention Law: Knowledge Loss Is Proportional to Representational Rotation in Fine-Tuned Language Models. With EgoRA: Entropy-Governed Orthogonality Regularization for Adaptation","display_name":"The Rotation-Retention Law: Knowledge Loss Is Proportional to Representational Rotation in Fine-Tuned Language Models. With EgoRA: Entropy-Governed Orthogonality Regularization for Adaptation","publication_year":2026,"publication_date":"2026-04-03","ids":{"openalex":"https://openalex.org/W7148635577","doi":"https://doi.org/10.5281/zenodo.19398708"},"language":"en","primary_location":{"id":"doi:10.5281/zenodo.19398708","doi":"https://doi.org/10.5281/zenodo.19398708","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.19398708","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article-journal"},"volume":null,"type":"article","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.5281/zenodo.19398708","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5132922089","display_name":"Mark Dillerop","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Dillerop, Mark","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5132922089"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.21537526,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9868000149726868,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.9868000149726868,"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/T10028","display_name":"Topic Modeling","score":0.01140000019222498,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.0003000000142492354,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/catastrophic-forgetting","display_name":"catastrophic forgetting","score":0.9760000109672546},{"id":"https://openalex.org/keywords/rotation-angle","display_name":"rotation angle","score":0.5360000133514404},{"id":"https://openalex.org/keywords/orthogonality-regularization","display_name":"orthogonality regularization","score":0.871999979019165},{"id":"https://openalex.org/keywords/parameter-efficient-fine-tuning","display_name":"parameter-efficient fine-tuning","score":0.8859999775886536},{"id":"https://openalex.org/keywords/lora","display_name":"LoRa","score":0.9520000219345093},{"id":"https://openalex.org/keywords/dora","display_name":"DORA","score":0.9829999804496765},{"id":"https://openalex.org/keywords/mmlu-benchmark","display_name":"MMLU benchmark","score":0.7160000205039978},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.5289999842643738}],"concepts":[{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.616599977016449},{"id":"https://openalex.org/C17137986","wikidata":"https://www.wikidata.org/wiki/Q215067","display_name":"Orthogonality","level":2,"score":0.5957000255584717},{"id":"https://openalex.org/C74050887","wikidata":"https://www.wikidata.org/wiki/Q848368","display_name":"Rotation (mathematics)","level":2,"score":0.5631999969482422},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.506600022315979},{"id":"https://openalex.org/C2776434776","wikidata":"https://www.wikidata.org/wiki/Q19246213","display_name":"Domain adaptation","level":3,"score":0.49709999561309814},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4602999985218048},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.4196999967098236},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.3921999931335449},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.3828999996185303},{"id":"https://openalex.org/C7149132","wikidata":"https://www.wikidata.org/wiki/Q1377840","display_name":"Forgetting","level":2,"score":0.36059999465942383},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.33959999680519104},{"id":"https://openalex.org/C2777027219","wikidata":"https://www.wikidata.org/wiki/Q1284190","display_name":"Constant (computer programming)","level":2,"score":0.334199994802475},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.31700000166893005},{"id":"https://openalex.org/C2781162219","wikidata":"https://www.wikidata.org/wiki/Q26250693","display_name":"Replicate","level":2,"score":0.3093000054359436},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.26980000734329224},{"id":"https://openalex.org/C148043351","wikidata":"https://www.wikidata.org/wiki/Q4456944","display_name":"Current (fluid)","level":2,"score":0.26759999990463257},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.266400009393692},{"id":"https://openalex.org/C107551265","wikidata":"https://www.wikidata.org/wiki/Q1458245","display_name":"Displacement (psychology)","level":2,"score":0.2653000056743622},{"id":"https://openalex.org/C4554734","wikidata":"https://www.wikidata.org/wiki/Q593744","display_name":"Knowledge base","level":2,"score":0.25870001316070557},{"id":"https://openalex.org/C176321772","wikidata":"https://www.wikidata.org/wiki/Q1430640","display_name":"Numerical stability","level":3,"score":0.25450000166893005},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.2529999911785126}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.5281/zenodo.19398708","doi":"https://doi.org/10.5281/zenodo.19398708","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.19398708","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article-journal"},{"id":"doi:10.5281/zenodo.19398709","doi":"https://doi.org/10.5281/zenodo.19398709","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.19398709","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article-journal"},{"id":"doi:10.5281/zenodo.19410504","doi":"https://doi.org/10.5281/zenodo.19410504","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.19410504","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article-journal"}],"best_oa_location":{"id":"doi:10.5281/zenodo.19398708","doi":"https://doi.org/10.5281/zenodo.19398708","is_oa":true,"landing_page_url":"https://doi.org/10.5281/zenodo.19398708","pdf_url":null,"source":{"id":"https://openalex.org/S4306400562","display_name":"Zenodo (CERN European Organization for Nuclear Research)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I67311998","host_organization_name":"European Organization for Nuclear Research","host_organization_lineage":["https://openalex.org/I67311998"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"article-journal"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[{"score":0.7911449670791626,"display_name":"Peace, justice, and strong institutions","id":"https://metadata.un.org/sdg/16"}],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"We":[0,22,77,103,254,292],"introduce":[1,41,225],"the":[2,13,57,65,79,88,98,107,119,128,201,218,256,270,335,342],"Rotation-Retention":[3,108,307],"Law:":[4,109],"knowledge":[5,83],"loss":[6,84,178,192,240],"during":[7,235],"fine-tuning":[8,73,160,323],"is":[9,81,85,118,127,276,334,354],"directly":[10],"proportional":[11,86],"to":[12,60,87,243,251,325,375],"mean":[14,89,129],"rotation":[15,90,130,234,265,272,286],"angle":[16,91,131],"of":[17,92,341],"attention":[18,93,134],"head":[19,94],"weight":[20,95],"vectors.":[21],"validate":[23,255],"this":[24,105,371],"geometric":[25,314],"relationship":[26],"across":[27,132,258],"20":[28,137],"experimental":[29,138],"conditions,":[30],"three":[31,144,153,165,259],"architectures":[32,145],"(Llama,":[33],"Mistral,":[34],"Phi-3),":[35],"and":[36,40,67,100,125,164,210,237,284,316,327],"multiple":[37],"model":[38,154],"scales,":[39],"EgoRA":[42,226,275],"\u2014":[43],"a":[44,298,313,318,357],"new":[45],"entropy-governed":[46],"orthogonality":[47],"regularization":[48],"method":[49],"that":[50,262],"substantially":[51],"reduces":[52,238],"catastrophic":[53,310],"forgetting.":[54],"I":[55],"invite":[56],"research":[58,343],"community":[59],"scrutinize":[61],"these":[62],"findings,":[63],"replicate":[64],"experiments,":[66],"provide":[68],"open":[69],"feedback.":[70],"===Why":[71],"does":[72],"destroy":[74],"pre-trained":[75,99],"knowledge?":[76],"show":[78],"answer":[80],"geometric:":[82],"vectors":[96],"between":[97],"adapted":[101],"states.":[102],"formalize":[104],"as":[106,297,312],"\u0394M":[110,117],"\u221d":[111],"\u03b8\u0304":[112,126,173,187],"=":[113,205,280],"E[arccos(cossim(w_i^base,":[114],"w_i^tuned))],":[115],"where":[116],"change":[120],"in":[121,182,196,217,370],"benchmark":[122,142],"performance":[123],"(MMLU)":[124],"all":[133,183,197],"heads.":[135],"Across":[136],"conditions":[139],"with":[140,214],"full":[141],"validation\u2014spanning":[143],"(Llama":[146],"3.1":[147],"8B,":[148],"Mistral":[149],"7B,":[150],"Phi-3":[151],"Mini),":[152],"scales":[155],"(1B,":[156],"3B,":[157],"8B),":[158],"two":[159],"domains":[161],"(instruction-following,":[162],"medical),":[163],"adaptation":[166],"methods":[167,368],"(LoRA,":[168],"DoRA,":[169],"EgoRA)\u2014we":[170],"find:":[171],"(i)":[172],">":[174],"5\u00b0":[175],"predicts":[176,190,288],"MMLU":[177,191,239,290],"exceeding":[179],"3":[180,194],"pp":[181,195],"high-rotation":[184,219],"conditions;":[185,199],"(ii)":[186],"<":[188,295],"3\u00b0":[189],"below":[193],"low-rotation":[198],"(iii)":[200],"proportionality":[202],"constant":[203],"k":[204],"|\u0394M|/\u03b8\u0304":[206],"varies":[207,266],"by":[208,241,267],"regime":[209],"scale":[211],"(0.22\u20130.96":[212],"pp/degree),":[213],"higher":[215,285],"stability":[216],"regime.":[220],"To":[221],"establish":[222],"causality,":[223],"we":[224],"(Entropy-Governed":[227],"Orthogonality":[228],"Regularization":[229],"for":[230,303,322,339],"Adaptation),":[231],"which":[232],"constrains":[233],"training":[236],"up":[242],"12.7\u00d7":[244],"while":[245,263],"reducing":[246],"damaged":[247],"heads":[248],"from":[249,274],"56%":[250],"under":[252,356],"4%.":[253],"law":[257],"architectures,":[260],"finding":[261],"absolute":[264],"6\u00d7":[268],"(3.6\u00b0\u201321.7\u00b0),":[269],"normalized":[271],"reduction":[273],"remarkably":[277],"consistent":[278],"(\u03b8\u0304_norm":[279],"0.42":[281],"\u00b1":[282],"0.06),":[283],"universally":[287],"greater":[289],"loss.":[291],"propose":[293],"\u03b8\u0304_norm":[294],"0.5":[296],"universal,":[299],"architecture-independent":[300],"certification":[301],"threshold":[302],"knowledge-preserving":[304],"fine-tuning.":[305],"The":[306,367],"Law":[308],"reframes":[309],"forgetting":[311],"phenomenon":[315],"provides":[317],"falsifiable,":[319],"architecture-agnostic":[320],"diagnostic":[321],"quality.Submitted":[324],"SSRN":[326],"Zenodo":[328],"on":[329],"April":[330],"3,":[331],"2026.":[332],"This":[333,352],"initial":[336],"public":[337],"release":[338],"publication":[340],"findings.":[344],"\u00a9":[345],"2026":[346],"Mark":[347],"Dillerop.":[348],"All":[349],"rights":[350],"reserved.":[351],"preprint":[353],"licensed":[355],"Creative":[358],"Commons":[359],"Attribution":[360],"4.0":[361],"International":[362],"License":[363],"(CC":[364],"BY":[365],"4.0).":[366],"described":[369],"work":[372],"are":[373],"subject":[374],"U.S.":[376],"Provisional":[377],"Patent":[378],"Application":[379],"No.":[380],"64/024,742.":[381]},"counts_by_year":[],"updated_date":"2026-10-08T09:06:37.294493","created_date":"2026-04-04T00:00:00"}