{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WCTPKLEPVYO2EKNPE2U57Y7T7T","short_pith_number":"pith:WCTPKLEP","schema_version":"1.0","canonical_sha256":"b0a6f52c8fae1da229af26a9dfe3f3fce01b323038dbc7f7f474dff65c60d716","source":{"kind":"arxiv","id":"2605.28567","version":1},"attestation_state":"computed","paper":{"title":"Semantic Optimal Transport for Sparse Autoencoder Feature Matching and Circuit Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"My T. Thai, Nguyen Do, Tue M. Cao","submitted_at":"2026-05-27T14:54:58Z","abstract_excerpt":"Sparse autoencoders (SAEs) have become a central tool for interpreting language models. However, two key SAE analyses that remain difficult to scale are (1) matching semantically similar features across multi-layers and (2) compressing large feature circuits into interpretable supernodes. Although these have been treated as separate problems, we show that both are instances of a more fundamental challenge, which we frame as the estimation of semantic distances between SAE features that lie on different activation manifolds. We introduce a distributional framework for this problem, in which eac"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.28567","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-27T14:54:58Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2e17122db56ddbf6224db644a95f5fcb8a7d4f42aae8ae7fd6c3e1a22c3cd08d","abstract_canon_sha256":"4819410bd4129b183ad0100412882565bfeb8fdfe081e39749f1712376bf1212"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T02:04:56.587540Z","signature_b64":"xcDwoESGOSE5jp2yHOf/vQHJiOBIm03jAATMBm1WK1Cjc6i/MwG7ugF6jT4GHxdee4MidZfmOi0Ru8JWwRacDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b0a6f52c8fae1da229af26a9dfe3f3fce01b323038dbc7f7f474dff65c60d716","last_reissued_at":"2026-05-28T02:04:56.587026Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T02:04:56.587026Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Semantic Optimal Transport for Sparse Autoencoder Feature Matching and Circuit Compression","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"My T. Thai, Nguyen Do, Tue M. Cao","submitted_at":"2026-05-27T14:54:58Z","abstract_excerpt":"Sparse autoencoders (SAEs) have become a central tool for interpreting language models. However, two key SAE analyses that remain difficult to scale are (1) matching semantically similar features across multi-layers and (2) compressing large feature circuits into interpretable supernodes. Although these have been treated as separate problems, we show that both are instances of a more fundamental challenge, which we frame as the estimation of semantic distances between SAE features that lie on different activation manifolds. We introduce a distributional framework for this problem, in which eac"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28567","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.28567/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2605.28567","created_at":"2026-05-28T02:04:56.587094+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28567v1","created_at":"2026-05-28T02:04:56.587094+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28567","created_at":"2026-05-28T02:04:56.587094+00:00"},{"alias_kind":"pith_short_12","alias_value":"WCTPKLEPVYO2","created_at":"2026-05-28T02:04:56.587094+00:00"},{"alias_kind":"pith_short_16","alias_value":"WCTPKLEPVYO2EKNP","created_at":"2026-05-28T02:04:56.587094+00:00"},{"alias_kind":"pith_short_8","alias_value":"WCTPKLEP","created_at":"2026-05-28T02:04:56.587094+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T","json":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T.json","graph_json":"https://pith.science/api/pith-number/WCTPKLEPVYO2EKNPE2U57Y7T7T/graph.json","events_json":"https://pith.science/api/pith-number/WCTPKLEPVYO2EKNPE2U57Y7T7T/events.json","paper":"https://pith.science/paper/WCTPKLEP"},"agent_actions":{"view_html":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T","download_json":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T.json","view_paper":"https://pith.science/paper/WCTPKLEP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28567&json=true","fetch_graph":"https://pith.science/api/pith-number/WCTPKLEPVYO2EKNPE2U57Y7T7T/graph.json","fetch_events":"https://pith.science/api/pith-number/WCTPKLEPVYO2EKNPE2U57Y7T7T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T/action/storage_attestation","attest_author":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T/action/author_attestation","sign_citation":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T/action/citation_signature","submit_replication":"https://pith.science/pith/WCTPKLEPVYO2EKNPE2U57Y7T7T/action/replication_record"}},"created_at":"2026-05-28T02:04:56.587094+00:00","updated_at":"2026-05-28T02:04:56.587094+00:00"}