{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:ACJQ7UNFZ3MSBW2RFHBMWFODW7","short_pith_number":"pith:ACJQ7UNF","schema_version":"1.0","canonical_sha256":"00930fd1a5ced920db5129c2cb15c3b7e9324cbf8592acfd5c116339fc79b5fb","source":{"kind":"arxiv","id":"2310.06301","version":1},"attestation_state":"computed","paper":{"title":"Dynamical versus Bayesian Phase Transitions in a Toy Model of Superposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Daniel Murfet, Edmund Lau, Jake Mendel, Susan Wei, Zhongtian Chen","submitted_at":"2023-10-10T04:26:04Z","abstract_excerpt":"We investigate phase transitions in a Toy Model of Superposition (TMS) using Singular Learning Theory (SLT). We derive a closed formula for the theoretical loss and, in the case of two hidden dimensions, discover that regular $k$-gons are critical points. We present supporting theory indicating that the local learning coefficient (a geometric invariant) of these $k$-gons determines phase transitions in the Bayesian posterior as a function of training sample size. We then show empirically that the same $k$-gon critical points also determine the behavior of SGD training. The picture that emerges"},"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":"2310.06301","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-10T04:26:04Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"fc59ccc08c3e600995b9f928bded22feed2ff0d36b4f9ba9f2965c45c1e93444","abstract_canon_sha256":"66528f1991226bbc74372552c8d48780f36bcb5c1d63bbdf59584066104395f9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:59:08.954270Z","signature_b64":"u9C57r5hIyf7vCNhbh7gDgnttzIe5ihdTYMmbuMhSlF9oXhQHilMPhOo+AZJqpJtXxKSnroGj+twGmI5AE1hDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"00930fd1a5ced920db5129c2cb15c3b7e9324cbf8592acfd5c116339fc79b5fb","last_reissued_at":"2026-07-05T06:59:08.953777Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:59:08.953777Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Dynamical versus Bayesian Phase Transitions in a Toy Model of Superposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Daniel Murfet, Edmund Lau, Jake Mendel, Susan Wei, Zhongtian Chen","submitted_at":"2023-10-10T04:26:04Z","abstract_excerpt":"We investigate phase transitions in a Toy Model of Superposition (TMS) using Singular Learning Theory (SLT). We derive a closed formula for the theoretical loss and, in the case of two hidden dimensions, discover that regular $k$-gons are critical points. We present supporting theory indicating that the local learning coefficient (a geometric invariant) of these $k$-gons determines phase transitions in the Bayesian posterior as a function of training sample size. We then show empirically that the same $k$-gon critical points also determine the behavior of SGD training. The picture that emerges"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.06301","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/2310.06301/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":"2310.06301","created_at":"2026-07-05T06:59:08.953838+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.06301v1","created_at":"2026-07-05T06:59:08.953838+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.06301","created_at":"2026-07-05T06:59:08.953838+00:00"},{"alias_kind":"pith_short_12","alias_value":"ACJQ7UNFZ3MS","created_at":"2026-07-05T06:59:08.953838+00:00"},{"alias_kind":"pith_short_16","alias_value":"ACJQ7UNFZ3MSBW2R","created_at":"2026-07-05T06:59:08.953838+00:00"},{"alias_kind":"pith_short_8","alias_value":"ACJQ7UNF","created_at":"2026-07-05T06:59:08.953838+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26620","citing_title":"Discovering Millions of Interpretable Features with Sparse Autoencoders","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05957","citing_title":"Dead Directions: Geometric Singular Learning","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2606.01302","citing_title":"Structure and Scale in Simplicial Sequence Modelling","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12809","citing_title":"Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces","ref_index":261,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7","json":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7.json","graph_json":"https://pith.science/api/pith-number/ACJQ7UNFZ3MSBW2RFHBMWFODW7/graph.json","events_json":"https://pith.science/api/pith-number/ACJQ7UNFZ3MSBW2RFHBMWFODW7/events.json","paper":"https://pith.science/paper/ACJQ7UNF"},"agent_actions":{"view_html":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7","download_json":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7.json","view_paper":"https://pith.science/paper/ACJQ7UNF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.06301&json=true","fetch_graph":"https://pith.science/api/pith-number/ACJQ7UNFZ3MSBW2RFHBMWFODW7/graph.json","fetch_events":"https://pith.science/api/pith-number/ACJQ7UNFZ3MSBW2RFHBMWFODW7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7/action/storage_attestation","attest_author":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7/action/author_attestation","sign_citation":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7/action/citation_signature","submit_replication":"https://pith.science/pith/ACJQ7UNFZ3MSBW2RFHBMWFODW7/action/replication_record"}},"created_at":"2026-07-05T06:59:08.953838+00:00","updated_at":"2026-07-05T06:59:08.953838+00:00"}