{"id":"https://openalex.org/W7167925788","doi":"https://doi.org/10.48550/arxiv.2607.08601","title":"It Takes a MAESTRO To Prune Bad Experts","display_name":"It Takes a MAESTRO To Prune Bad Experts","publication_year":2026,"publication_date":"2026-07-09","ids":{"openalex":"https://openalex.org/W7167925788","doi":"https://doi.org/10.48550/arxiv.2607.08601"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2607.08601","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.08601","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,"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.2607.08601","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5140403131","display_name":"Palaash Goel","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Goel, Palaash","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5002397691","display_name":"Ayush Maheshwari","orcid":"https://orcid.org/0000-0002-2542-3750"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Maheshwari, Ayush","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5140445599","display_name":"Tanmoy Chakraborty","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chakraborty, Tanmoy","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/T11896","display_name":"Opportunistic and Delay-Tolerant Networks","score":0.18479999899864197,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T11896","display_name":"Opportunistic and Delay-Tolerant Networks","score":0.18479999899864197,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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/T11704","display_name":"Mobile Crowdsensing and Crowdsourcing","score":0.11069999635219574,"subfield":{"id":"https://openalex.org/subfields/1706","display_name":"Computer Science Applications"},"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.04439999908208847,"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/pruning","display_name":"Pruning","score":0.6658999919891357},{"id":"https://openalex.org/keywords/heuristics","display_name":"Heuristics","score":0.5964999794960022},{"id":"https://openalex.org/keywords/markov-chain","display_name":"Markov chain","score":0.5080999732017517},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5059999823570251},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.47440001368522644},{"id":"https://openalex.org/keywords/encode","display_name":"ENCODE","score":0.4101000130176544},{"id":"https://openalex.org/keywords/forgetting","display_name":"Forgetting","score":0.4101000130176544},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.39959999918937683},{"id":"https://openalex.org/keywords/expert-system","display_name":"Expert system","score":0.39739999175071716}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7042999863624573},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6658999919891357},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.5964999794960022},{"id":"https://openalex.org/C98763669","wikidata":"https://www.wikidata.org/wiki/Q176645","display_name":"Markov chain","level":2,"score":0.5080999732017517},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5059999823570251},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4871000051498413},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.47440001368522644},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4530999958515167},{"id":"https://openalex.org/C7149132","wikidata":"https://www.wikidata.org/wiki/Q1377840","display_name":"Forgetting","level":2,"score":0.4101000130176544},{"id":"https://openalex.org/C66746571","wikidata":"https://www.wikidata.org/wiki/Q1134833","display_name":"ENCODE","level":3,"score":0.4101000130176544},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.39959999918937683},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.39739999175071716},{"id":"https://openalex.org/C122044880","wikidata":"https://www.wikidata.org/wiki/Q5498822","display_name":"Ergodic theory","level":2,"score":0.3693999946117401},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.3626999855041504},{"id":"https://openalex.org/C159886148","wikidata":"https://www.wikidata.org/wiki/Q176645","display_name":"Markov process","level":2,"score":0.3301999866962433},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.3160000145435333},{"id":"https://openalex.org/C2780992000","wikidata":"https://www.wikidata.org/wiki/Q17016113","display_name":"Generator (circuit theory)","level":3,"score":0.3025999963283539},{"id":"https://openalex.org/C111350023","wikidata":"https://www.wikidata.org/wiki/Q1191869","display_name":"Markov chain Monte Carlo","level":3,"score":0.29739999771118164},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.2964000105857849},{"id":"https://openalex.org/C185874996","wikidata":"https://www.wikidata.org/wiki/Q269699","display_name":"Interdependence","level":2,"score":0.2930999994277954},{"id":"https://openalex.org/C163836022","wikidata":"https://www.wikidata.org/wiki/Q6771326","display_name":"Markov model","level":3,"score":0.2824000120162964},{"id":"https://openalex.org/C136197465","wikidata":"https://www.wikidata.org/wiki/Q1729295","display_name":"Variety (cybernetics)","level":2,"score":0.28119999170303345},{"id":"https://openalex.org/C105002631","wikidata":"https://www.wikidata.org/wiki/Q4833645","display_name":"Subject-matter expert","level":3,"score":0.2793999910354614},{"id":"https://openalex.org/C177769412","wikidata":"https://www.wikidata.org/wiki/Q278090","display_name":"Prior probability","level":3,"score":0.2782000005245209},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.27000001072883606}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2607.08601","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.08601","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,"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.2607.08601","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2607.08601","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,"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":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Sparsely-activated":[0],"Mixture-of-Experts":[1],"(MoE)":[2],"language":[3],"models":[4,80,141],"achieve":[5],"remarkable":[6],"inference":[7],"efficiency":[8],"by":[9,115],"activating":[10],"only":[11],"a":[12,31,71,96,124],"small":[13],"fraction":[14],"of":[15,58],"parameters":[16],"per":[17],"token,":[18],"yet":[19],"their":[20],"full":[21],"expert":[22,45,82],"banks":[23],"reside":[24],"in":[25,119],"memory":[26],"at":[27],"all":[28],"times,":[29],"creating":[30],"prohibitive":[32],"deployment":[33],"bottleneck.":[34],"Existing":[35],"structured":[36,72],"pruning":[37,73,139],"methods,":[38],"largely":[39],"designed":[40,75],"for":[41,76],"dense":[42],"transformers,":[43],"assess":[44],"importance":[46,99],"using":[47],"locally":[48],"derived":[49],"heuristics":[50],"that":[51,79,136,142],"are":[52],"blind":[53],"to":[54,117],"the":[55],"interdependent":[56],"nature":[57],"MoE":[59,77],"routing.":[60],"We":[61],"introduce":[62],"MAESTRO":[63,111],"(Markov-chain":[64],"Approximated":[65],"Expert":[66],"Sparsification":[67],"via":[68],"Transition-based":[69],"ROuting),":[70],"framework":[74],"architectures":[78],"autoregressive":[81],"activation":[83],"trajectories":[84],"as":[85],"Ergodic":[86],"Markov":[87],"chains":[88],"whose":[89],"stationary":[90],"distributions":[91],"encode":[92],"cross-layer":[93],"dependencies,":[94],"yielding":[95],"globally":[97],"aware":[98],"heuristic.":[100],"Evaluated":[101],"across":[102,146],"five":[103],"diverse":[104],"domains":[105],"including":[106],"Safety,":[107],"Bias,":[108],"and":[109],"Ethics,":[110],"outperforms":[112],"state-of-the-art":[113],"baselines":[114],"up":[116],"10.61%":[118],"average":[120],"performance":[121],"retention":[122],"under":[123],"strict":[125],"50%":[126],"compression":[127],"regime,":[128],"while":[129],"exhibiting":[130],"substantially":[131],"lower":[132],"cross-task":[133],"variance,":[134],"indicating":[135],"global,":[137],"routing-congruent":[138],"produces":[140],"generalize":[143],"more":[144],"consistently":[145],"heterogeneous":[147],"tasks.":[148]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-07-11T00:00:00"}