{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:IWPYLERGCBVQK4KIMNIRTU3MBX","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"1c7b5a1f077e465aafbf6c1cd9cbf0b0269a3e935b4e23eb801430934d259132","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-01-31T22:47:54Z","title_canon_sha256":"bfa5a1126602dbd2b9a2d6632a4e83070f74210d3a43e7e5451ab40fd5551906"},"schema_version":"1.0","source":{"id":"2602.00924","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.00924","created_at":"2026-05-20T00:04:23Z"},{"alias_kind":"arxiv_version","alias_value":"2602.00924v3","created_at":"2026-05-20T00:04:23Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.00924","created_at":"2026-05-20T00:04:23Z"},{"alias_kind":"pith_short_12","alias_value":"IWPYLERGCBVQ","created_at":"2026-05-20T00:04:23Z"},{"alias_kind":"pith_short_16","alias_value":"IWPYLERGCBVQK4KI","created_at":"2026-05-20T00:04:23Z"},{"alias_kind":"pith_short_8","alias_value":"IWPYLERG","created_at":"2026-05-20T00:04:23Z"}],"graph_snapshots":[{"event_id":"sha256:d4aeaa7b28f1a7aaaef07c26374bdebb425f09b72eec724242ebb642e45e321f","target":"graph","created_at":"2026-05-20T00:04:23Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Validated on Stable Diffusion 3.5, our approach demonstrates compositional generalization, successfully reconstructing images with concept combinations unseen during training, and enabling feature-level intervention for semantic image editing without prompt modification."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That jointly learning sparse concept embeddings with decoder weights will produce features that align with human semantics and support true compositional generalization without overfitting to the supervised training concepts or losing reconstruction fidelity."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Supervised SAEs using concept embeddings from neural collapse theory achieve compositional generalization on unseen concept combinations in Stable Diffusion 3.5 for reconstruction and feature-level image editing."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Supervised sparse auto-encoders achieve compositional generalization by jointly learning sparse concept embeddings and decoder weights."}],"snapshot_sha256":"87682f916ac9f2921548d83a871e1e44f9e459bdff5224d616d15e6d28caea51"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"65f36526cffe216f2b0909173be2b0e5a31ac9b476fb2f59f33728cf0b0e2950"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2602.00924/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Sparse auto-encoders (SAEs) have re-emerged as a prominent method for mechanistic interpretability, yet they face two significant challenges: the non-smoothness of the $L_1$ penalty, which hinders reconstruction and scalability, and a lack of alignment between learned features and human semantics. In this paper, we address these limitations by adapting unconstrained feature models, a mathematical framework from neural collapse theory, and by supervising the task. We supervise (decoder-only) SAEs to reconstruct feature vectors by jointly learning sparse concept embeddings and decoder weights. V","authors_text":"Haixuan Xavier Tao, Hugo Wallner, Ouns El Harzli, Yoonsoo Nam","cross_cats":[],"headline":"Supervised sparse auto-encoders achieve compositional generalization by jointly learning sparse concept embeddings and decoder weights.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-01-31T22:47:54Z","title":"Supervised sparse auto-encoders for interpretable and compositional representations"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.00924","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-16T08:29:02.971840Z","id":"9b2c23e3-def3-4c48-8304-ba62599a5e32","model_set":{"reader":"grok-4.3"},"one_line_summary":"Supervised SAEs using concept embeddings from neural collapse theory achieve compositional generalization on unseen concept combinations in Stable Diffusion 3.5 for reconstruction and feature-level image editing.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Supervised sparse auto-encoders achieve compositional generalization by jointly learning sparse concept embeddings and decoder weights.","strongest_claim":"Validated on Stable Diffusion 3.5, our approach demonstrates compositional generalization, successfully reconstructing images with concept combinations unseen during training, and enabling feature-level intervention for semantic image editing without prompt modification.","weakest_assumption":"That jointly learning sparse concept embeddings with decoder weights will produce features that align with human semantics and support true compositional generalization without overfitting to the supervised training concepts or losing reconstruction fidelity."}},"verdict_id":"9b2c23e3-def3-4c48-8304-ba62599a5e32"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f65230655e836c0d4c3e5614b5ecc9538a4e5d9aa9cde735ce3006e8414ed716","target":"record","created_at":"2026-05-20T00:04:23Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"1c7b5a1f077e465aafbf6c1cd9cbf0b0269a3e935b4e23eb801430934d259132","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-01-31T22:47:54Z","title_canon_sha256":"bfa5a1126602dbd2b9a2d6632a4e83070f74210d3a43e7e5451ab40fd5551906"},"schema_version":"1.0","source":{"id":"2602.00924","kind":"arxiv","version":3}},"canonical_sha256":"459f859226106b057148635119d36c0dd29b1e7db78a291333705dceef851ba1","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"459f859226106b057148635119d36c0dd29b1e7db78a291333705dceef851ba1","first_computed_at":"2026-05-20T00:04:23.959441Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:04:23.959441Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"7y7fH2Ewmh0rVPmBrmklEW7sb0ZQUjX981N3fFRXYrDcnUPavmcVnFLJfWSg+mmX3El0TdmlQ3gfScunIL83CA==","signature_status":"signed_v1","signed_at":"2026-05-20T00:04:23.960252Z","signed_message":"canonical_sha256_bytes"},"source_id":"2602.00924","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f65230655e836c0d4c3e5614b5ecc9538a4e5d9aa9cde735ce3006e8414ed716","sha256:d4aeaa7b28f1a7aaaef07c26374bdebb425f09b72eec724242ebb642e45e321f"],"state_sha256":"1a16ec5ee32b70247d8cf25c5d4afd5f2ebb6a936089b9433f12cb6d617ca319"}