{"id":"https://openalex.org/W7162115752","doi":"https://doi.org/10.48550/arxiv.2605.22658","title":"SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation","display_name":"SegCompass: Exploring Interpretable Alignment with Sparse Autoencoders for Enhanced Reasoning Segmentation","publication_year":2026,"publication_date":"2026-05-21","ids":{"openalex":"https://openalex.org/W7162115752","doi":"https://doi.org/10.48550/arxiv.2605.22658"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.22658","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.22658","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.22658","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136776401","display_name":"Zhenyu Lu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lu, Zhenyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136782134","display_name":"Liupeng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Liupeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136771663","display_name":"Jinpeng Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Jinpeng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136766297","display_name":"Haoqian Kang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kang, Haoqian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136801238","display_name":"Yan Feng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Feng, Yan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136734628","display_name":"Ke Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Ke","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136786015","display_name":"Yaowei Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yaowei","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9672999978065491,"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"}},"topics":[{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9672999978065491,"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"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.00839999970048666,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.005499999970197678,"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/interpretability","display_name":"Interpretability","score":0.8454999923706055},{"id":"https://openalex.org/keywords/autoencoder","display_name":"Autoencoder","score":0.5956000089645386},{"id":"https://openalex.org/keywords/segmentation","display_name":"Segmentation","score":0.5194000005722046},{"id":"https://openalex.org/keywords/visual-reasoning","display_name":"Visual reasoning","score":0.48649999499320984},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.48579999804496765},{"id":"https://openalex.org/keywords/path","display_name":"Path (computing)","score":0.4643000066280365},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.41370001435279846},{"id":"https://openalex.org/keywords/encoder","display_name":"Encoder","score":0.41190001368522644},{"id":"https://openalex.org/keywords/salient","display_name":"Salient","score":0.40310001373291016}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.8454999923706055},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7616000175476074},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6960999965667725},{"id":"https://openalex.org/C101738243","wikidata":"https://www.wikidata.org/wiki/Q786435","display_name":"Autoencoder","level":3,"score":0.5956000089645386},{"id":"https://openalex.org/C89600930","wikidata":"https://www.wikidata.org/wiki/Q1423946","display_name":"Segmentation","level":2,"score":0.5194000005722046},{"id":"https://openalex.org/C2777508537","wikidata":"https://www.wikidata.org/wiki/Q7936620","display_name":"Visual reasoning","level":2,"score":0.48649999499320984},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.48579999804496765},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.4643000066280365},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.41370001435279846},{"id":"https://openalex.org/C118505674","wikidata":"https://www.wikidata.org/wiki/Q42586063","display_name":"Encoder","level":2,"score":0.41190001368522644},{"id":"https://openalex.org/C2780719617","wikidata":"https://www.wikidata.org/wiki/Q1030752","display_name":"Salient","level":2,"score":0.40310001373291016},{"id":"https://openalex.org/C43521106","wikidata":"https://www.wikidata.org/wiki/Q2165493","display_name":"Pipeline (software)","level":2,"score":0.36579999327659607},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3366999924182892},{"id":"https://openalex.org/C77637269","wikidata":"https://www.wikidata.org/wiki/Q7002051","display_name":"Neural coding","level":2,"score":0.3312000036239624},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.32829999923706055},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.32710000872612},{"id":"https://openalex.org/C162307627","wikidata":"https://www.wikidata.org/wiki/Q204833","display_name":"Enhanced Data Rates for GSM Evolution","level":2,"score":0.31790000200271606},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.3077000081539154},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.30469998717308044},{"id":"https://openalex.org/C100776233","wikidata":"https://www.wikidata.org/wiki/Q2532492","display_name":"Bridge (graph theory)","level":2,"score":0.301800012588501},{"id":"https://openalex.org/C204241405","wikidata":"https://www.wikidata.org/wiki/Q461499","display_name":"Transformation (genetics)","level":3,"score":0.29660001397132874},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.29030001163482666},{"id":"https://openalex.org/C124504099","wikidata":"https://www.wikidata.org/wiki/Q56933","display_name":"Image segmentation","level":3,"score":0.27399998903274536},{"id":"https://openalex.org/C34413123","wikidata":"https://www.wikidata.org/wiki/Q170978","display_name":"Robotics","level":3,"score":0.26930001378059387},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.2651999890804291},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.2628999948501587},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.25949999690055847},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.25290000438690186},{"id":"https://openalex.org/C127759330","wikidata":"https://www.wikidata.org/wiki/Q637416","display_name":"Codebook","level":2,"score":0.2517000138759613}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.22658","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.22658","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.22658","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.22658","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":"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"},"sustainable_development_goals":[{"display_name":"Quality Education","score":0.43459659814834595,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"While":[0],"large":[1],"language":[2],"models":[3],"provide":[4],"strong":[5,199],"compositional":[6],"reasoning,":[7],"existing":[8],"reasoning":[9,17,150],"segmentation":[10,154],"pipelines":[11],"fail":[12],"to":[13,18,68],"transparently":[14],"connect":[15],"this":[16,53,118],"visual":[19,102,193],"perception.":[20],"Current":[21],"methods,":[22],"such":[23],"as":[24,46],"latent":[25,169],"query":[26,112],"alignment,":[27],"are":[28,121],"end-to-end":[29,60],"yet":[30],"opaque":[31],"\"black":[32],"boxes\".":[33],"Conversely,":[34],"textual":[35,175],"localization":[36],"readout":[37],"is":[38,93,142,164,226],"merely":[39],"readable,":[40],"not":[41],"truly":[42],"interpretable,":[43,72],"often":[44],"functioning":[45],"an":[47,59,70,78,94],"unconstrained":[48],"post-hoc":[49],"step.":[50],"To":[51],"bridge":[52],"interpretability":[54],"gap,":[55],"we":[56],"propose":[57],"SegCompass,":[58],"model":[61,141],"that":[62,96,133,163,184,214],"leverages":[63],"a":[64,84,105,126,130,160,198],"Sparse":[65],"Autoencoder":[66],"(SAE)":[67],"forge":[69],"explicit,":[71],"and":[73,101,171,194,209,222],"differentiable":[74],"alignment":[75],"pathway.":[76],"Given":[77],"image-instruction":[79],"pair,":[80],"SegCompass":[81,185,215],"first":[82],"generates":[83],"chain-of-thought":[85],"(CoT)":[86],"trace.":[87],"The":[88,139],"core":[89],"of":[90,204],"our":[91,192],"method":[92],"SAE":[95],"maps":[97],"both":[98],"the":[99,135,149,202,205],"CoT":[100],"tokens":[103],"into":[104,129],"shared,":[106],"high-dimensional":[107],"sparse":[108,207],"concept":[109],"space.":[110],"A":[111],"codebook":[113],"selects":[114],"salient":[115],"concepts":[116,208],"from":[117],"space,":[119],"which":[120],"then":[122],"spatially":[123],"grounded":[124],"by":[125],"slot":[127],"mapper":[128],"multi-slot":[131],"heatmap":[132],"guides":[134],"final":[136,210],"mask":[137,211],"decoder.":[138],"entire":[140],"trained":[143],"jointly,":[144],"unifying":[145],"reinforcement":[146],"learning":[147],"for":[148],"path":[151],"with":[152],"standard":[153],"supervision.":[155],"This":[156],"SAE-driven":[157],"interface":[158],"provides":[159],"\"white-box\"":[161],"connection":[162],"significantly":[165],"more":[166,172],"traceable":[167],"than":[168,174],"queries":[170],"coherent":[173],"readouts.":[176],"Extensive":[177],"experiments":[178],"on":[179],"five":[180],"challenging":[181],"benchmarks":[182],"demonstrate":[183],"matches":[186],"or":[187],"surpasses":[188],"state-of-the-art":[189],"performance.":[190],"Crucially,":[191],"quantitative":[195],"analyses":[196],"show":[197],"correlation":[200],"between":[201],"quality":[203],"learned":[206],"accuracy,":[212],"confirming":[213],"achieves":[216],"superior":[217],"results":[218],"through":[219],"its":[220],"enhanced":[221],"inspectable":[223],"alignment.":[224],"Code":[225],"available":[227],"at":[228],"https://github.com/ZhenyuLU-Heliodore/SegCompass.":[229]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-23T00:00:00"}