{"id":"https://openalex.org/W4400375635","doi":"https://doi.org/10.48550/arxiv.2407.02646","title":"A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models","display_name":"A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models","publication_year":2024,"publication_date":"2024-07-02","ids":{"openalex":"https://openalex.org/W4400375635","doi":"https://doi.org/10.48550/arxiv.2407.02646"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2407.02646","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.02646","pdf_url":"https://arxiv.org/pdf/2407.02646","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"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2407.02646","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5033331273","display_name":"Daking Rai","orcid":"https://orcid.org/0009-0003-3769-4974","observed_orcids":["https://orcid.org/0009-0003-3769-4974"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rai, Daking","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5074749353","display_name":"Yilun Zhou","orcid":"https://orcid.org/0000-0002-7085-3880","observed_orcids":["https://orcid.org/0000-0002-7085-3880"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Yilun","raw_affiliation_strings":["Datadog AI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Datadog AI Research","institution_ids":[]}]},{"author_position":"middle","author":{"id":null,"display_name":"Feng, Shi","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Shi","raw_affiliation_strings":["Datadog AI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Datadog AI Research","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5062813195","display_name":"Abulhair Saparov","orcid":"https://orcid.org/0000-0002-0450-9125","observed_orcids":["https://orcid.org/0000-0002-0450-9125"]},"institutions":[{"id":"https://openalex.org/I193531525","display_name":"George Washington University","ror":"https://ror.org/00y4zzh67","country_code":"US","type":"education","lineage":["https://openalex.org/I193531525"]},{"id":"https://openalex.org/I219193219","display_name":"Purdue University West Lafayette","ror":"https://ror.org/02dqehb95","country_code":"US","type":"education","lineage":["https://openalex.org/I219193219"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Saparov, Abulhair","raw_affiliation_strings":["George Washington University","Purdue University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"George Washington University","institution_ids":["https://openalex.org/I193531525"]},{"raw_affiliation_string":"Purdue University","institution_ids":["https://openalex.org/I219193219"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101616607","display_name":"Ziyu Yao","orcid":"https://orcid.org/0009-0007-4571-3505","observed_orcids":["https://orcid.org/0009-0007-4571-3505"]},"institutions":[{"id":"https://openalex.org/I219193219","display_name":"Purdue University West Lafayette","ror":"https://ror.org/02dqehb95","country_code":"US","type":"education","lineage":["https://openalex.org/I219193219"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yao, Ziyu","raw_affiliation_strings":["Purdue University","University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Purdue University","institution_ids":["https://openalex.org/I219193219"]},{"raw_affiliation_string":"University","institution_ids":[]}]}],"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":13,"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":1.0,"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":1.0,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.0,"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":0.0,"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":0.9950000047683716},{"id":"https://openalex.org/keywords/transformer-based-language-models","display_name":"transformer language models","score":0.8349999785423279},{"id":"https://openalex.org/keywords/neural-network-interpretability","display_name":"neural network interpretability","score":0.8029999732971191},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.5740000009536743}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9094201922416687},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.599858283996582},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5157299637794495},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3212231397628784},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.28908273577690125},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.1700969934463501},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.07706466317176819},{"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:2407.02646","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.02646","pdf_url":"https://arxiv.org/pdf/2407.02646","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.2407.02646","doi":"https://doi.org/10.48550/arxiv.2407.02646","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2407.02646","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:2407.02646","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.02646","pdf_url":"https://arxiv.org/pdf/2407.02646","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":[{"id":"https://openalex.org/G1129185673","display_name":"Collaborative Research: SHF: Medium: Toward Understandability and Interpretability for Neural Language Models of Source Code","funder_award_id":"2423813","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G2913428378","display_name":"Collaborative Research: SHF: Medium: Toward Understandability and Interpretability for Neural Language Models of Source Code","funder_award_id":"2311468","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320309029","display_name":"George Mason University","ror":"https://ror.org/02jqj7156"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4400375635.pdf","grobid_xml":"https://content.openalex.org/works/W4400375635.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7211958469","https://openalex.org/W4414792191","https://openalex.org/W7164409677","https://openalex.org/W7170177632","https://openalex.org/W7125680013","https://openalex.org/W7140121720","https://openalex.org/W7125584057","https://openalex.org/W4403090414","https://openalex.org/W4399038319","https://openalex.org/W7202232122"],"abstract_inverted_index":{"Mechanistic":[0],"interpretability":[1,8],"(MI)":[2],"is":[3],"an":[4],"emerging":[5],"sub-field":[6],"of":[7,81,95],"that":[9,49],"seeks":[10],"to":[11,62,126],"understand":[12],"a":[13,58,71,75,118,122],"neural":[14],"network":[15],"model":[16],"by":[17,130],"reverse-engineering":[18],"its":[19],"internal":[20],"computations.":[21],"Recently,":[22],"MI":[23,82,145,164],"has":[24,45],"garnered":[25],"significant":[26],"attention":[27],"for":[28,60,108,124,146,163],"interpreting":[29],"transformer-based":[30],"language":[31],"models":[32],"(LMs),":[33],"resulting":[34],"in":[35,97,111,137,155],"many":[36],"novel":[37],"insights":[38,53],"yet":[39],"introducing":[40],"new":[41],"challenges.":[42],"However,":[43],"there":[44],"not":[46],"been":[47],"work":[48],"comprehensively":[50],"reviews":[51],"these":[52],"and":[54,105,143,158],"challenges,":[55],"particularly":[56],"as":[57,121],"guide":[59],"newcomers":[61],"this":[63,67],"field.":[64],"To":[65],"fill":[66],"gap,":[68],"we":[69,116,150],"provide":[70],"comprehensive":[72],"survey":[73],"from":[74],"task-centric":[76,119],"perspective,":[77],"organizing":[78],"the":[79,92,101,112,128,152,156],"taxonomy":[80,120],"research":[83,86],"around":[84],"specific":[85],"questions":[87],"or":[88],"tasks.":[89],"We":[90],"outline":[91],"fundamental":[93],"objects":[94],"study":[96],"MI,":[98],"along":[99],"with":[100],"techniques,":[102],"evaluation":[103],"methods,":[104],"key":[106],"findings":[107],"each":[109],"task":[110],"taxonomy.":[113],"In":[114],"particular,":[115],"present":[117],"roadmap":[123],"beginners":[125],"navigate":[127],"field":[129,157],"helping":[131],"them":[132],"quickly":[133],"identify":[134],"impactful":[135],"problems":[136],"which":[138],"they":[139],"are":[140],"most":[141],"interested":[142],"leverage":[144],"their":[147],"benefit.":[148],"Finally,":[149],"discuss":[151],"current":[153],"gaps":[154],"suggest":[159],"potential":[160],"future":[161],"directions":[162],"research.":[165]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":9},{"year":2024,"cited_by_count":1}],"updated_date":"2026-11-03T10:50:19.383937","created_date":"2024-07-06T00:00:00"}