{"id":"https://openalex.org/W4407759155","doi":"https://doi.org/10.48550/arxiv.2502.12131","title":"Transformer Dynamics: A neuroscientific approach to interpretability of large language models","display_name":"Transformer Dynamics: A neuroscientific approach to interpretability of large language models","publication_year":2025,"publication_date":"2025-02-17","ids":{"openalex":"https://openalex.org/W4407759155","doi":"https://doi.org/10.48550/arxiv.2502.12131"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2502.12131","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.12131","pdf_url":"https://arxiv.org/pdf/2502.12131","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"parent_works":[],"lineage":["https://openalex.org/W4407759155"],"child_works_count":0,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2502.12131","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5089995448","display_name":"Jesseba Fernando","orcid":"https://orcid.org/0000-0001-6776-947X","observed_orcids":["https://orcid.org/0000-0001-6776-947X"]},"institutions":[{"id":"https://openalex.org/I12912129","display_name":"Northeastern University","ror":"https://ror.org/04t5xt781","country_code":"US","type":"education","lineage":["https://openalex.org/I12912129"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Fernando, Jesseba","raw_affiliation_strings":["Network Science Institute , Northeastern University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Network Science Institute , Northeastern University","institution_ids":["https://openalex.org/I12912129"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5089601338","display_name":"Grigori Guitchounts","orcid":"https://orcid.org/0000-0003-0142-0149","observed_orcids":["https://orcid.org/0000-0003-0142-0149"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guitchounts, Grigori","raw_affiliation_strings":["Independent"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Independent","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":16,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.9965999722480774,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.9965999722480774,"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/T10320","display_name":"Neural Networks and Applications","score":0.0013000000035390258,"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":0.0006000000284984708,"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/mechanistic-interpretability","display_name":"mechanistic interpretability","score":0.8939999938011169},{"id":"https://openalex.org/keywords/dynamical-systems-theory","display_name":"dynamical systems theory","score":0.8539999723434448},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7699999809265137},{"id":"https://openalex.org/keywords/periodic-orbits","display_name":"periodic orbits","score":0.7160000205039978},{"id":"https://openalex.org/keywords/attractor-dynamics","display_name":"attractor dynamics","score":0.9399999976158142},{"id":"https://openalex.org/keywords/deep-learning-interpretability","display_name":"deep learning interpretability","score":0.3269999921321869}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9370110034942627},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.548017144203186},{"id":"https://openalex.org/C145912823","wikidata":"https://www.wikidata.org/wiki/Q113558","display_name":"Dynamics (music)","level":2,"score":0.4880203306674957},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.4610516130924225},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4444708228111267},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.3754749894142151},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.34483784437179565},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3403123915195465},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.32584118843078613},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.11616382002830505},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.0651562511920929},{"id":"https://openalex.org/C19417346","wikidata":"https://www.wikidata.org/wiki/Q7922","display_name":"Pedagogy","level":1,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2502.12131","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.12131","pdf_url":"https://arxiv.org/pdf/2502.12131","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2502.12131","doi":"https://doi.org/10.48550/arxiv.2502.12131","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2502.12131","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2502.12131","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.12131","pdf_url":"https://arxiv.org/pdf/2502.12131","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4407759155.pdf","grobid_xml":"https://content.openalex.org/works/W4407759155.grobid-xml"},"referenced_works_count":12,"referenced_works":["https://openalex.org/W2135993484","https://openalex.org/W2462906003","https://openalex.org/W2948981900","https://openalex.org/W2955382287","https://openalex.org/W3041725488","https://openalex.org/W4388424695","https://openalex.org/W4394737082","https://openalex.org/W4395443869","https://openalex.org/W4400065290","https://openalex.org/W4403321836","https://openalex.org/W4406072134","https://openalex.org/W4411881822"],"related_works":["https://openalex.org/W4407385206","https://openalex.org/W7161247731","https://openalex.org/W7125278936","https://openalex.org/W4414759147","https://openalex.org/W4399655876","https://openalex.org/W7160231192","https://openalex.org/W7165356661","https://openalex.org/W7143398690","https://openalex.org/W7077041677","https://openalex.org/W3194564849"],"abstract_inverted_index":{"As":[0],"artificial":[1],"intelligence":[2],"models":[3],"have":[4],"exploded":[5],"in":[6,27,38,49,82,110],"scale":[7],"and":[8,86,120],"capability,":[9],"understanding":[10,142],"of":[11,23,65,129,143],"their":[12],"internal":[13],"mechanisms":[14],"remains":[15],"a":[16,32,55,78,104,124,127],"critical":[17],"challenge.":[18],"Inspired":[19],"by":[20],"the":[21,45,75,83,101,111],"success":[22],"dynamical":[24,56,117],"systems":[25,118],"approaches":[26],"neuroscience,":[28],"here":[29],"we":[30],"propose":[31],"novel":[33],"framework":[34],"for":[35,126],"studying":[36],"computations":[37],"deep":[39],"learning":[40],"systems.":[41],"We":[42,61],"focus":[43],"on":[44],"residual":[46],"stream":[47],"(RS)":[48],"transformer":[50],"models,":[51],"conceptualizing":[52],"it":[53],"as":[54],"system":[57],"evolving":[58],"across":[59,72],"layers.":[60,113],"find":[62],"that":[63,131],"activations":[64],"individual":[66,92],"RS":[67,76,84,102],"units":[68,93],"exhibit":[69],"strong":[70],"continuity":[71],"layers,":[73,90],"despite":[74],"being":[77],"non-privileged":[79],"basis.":[80],"Activations":[81],"accelerate":[85],"grow":[87],"denser":[88],"over":[89],"while":[91],"trace":[94],"unstable":[95],"periodic":[96],"orbits.":[97],"In":[98],"reduced-dimensional":[99],"spaces,":[100],"follows":[103],"curved":[105],"trajectory":[106],"with":[107,135],"attractor-like":[108],"dynamics":[109],"lower":[112],"These":[114],"insights":[115],"bridge":[116],"theory":[119],"mechanistic":[121],"interpretability,":[122],"establishing":[123],"foundation":[125],"\"neuroscience":[128],"AI\"":[130],"combines":[132],"theoretical":[133],"rigor":[134],"large-scale":[136],"data":[137],"analysis":[138],"to":[139],"advance":[140],"our":[141],"modern":[144],"neural":[145],"networks.":[146]},"counts_by_year":[{"year":2026,"cited_by_count":12},{"year":2025,"cited_by_count":4}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2025-02-20T00:00:00"}