{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:VS3FT6ZZ2KRY7TWNPVLLWUVKXP","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":"07849c9d1a4bf545b3e2a82e4b983491ba8c311b3d8a1b2b18df2693d350a8f2","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T03:35:46Z","title_canon_sha256":"053ff61afba5cb9c959c212834e1d656d977b42a269996ce53447947c04bb37a"},"schema_version":"1.0","source":{"id":"2605.14323","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14323","created_at":"2026-05-17T23:39:09Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14323v1","created_at":"2026-05-17T23:39:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14323","created_at":"2026-05-17T23:39:09Z"},{"alias_kind":"pith_short_12","alias_value":"VS3FT6ZZ2KRY","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"VS3FT6ZZ2KRY7TWN","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"VS3FT6ZZ","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:015b457a7ca8f04067702ce48f7ec3781377c1dfc302a45def00fccd85430c7a","target":"graph","created_at":"2026-05-17T23:39:09Z","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":"In low-data fine-tuning settings, DLR matches or outperforms supervised fine-tuning across four datasets and six models, achieving a mean gain of +6.6 percentage points, while prior discrete-latent baselines consistently underperform SFT."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the optimality guarantees and search principle from General Dijkstra Search in MDPs transfer effectively to the non-stationary, high-dimensional setting of language model post-training without introducing hidden biases or optimization instabilities."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Dynamic Latent Routing jointly learns discrete latent codes, routing policies, and model parameters via dynamic search to match or exceed supervised fine-tuning by 6.6 points on average in low-data settings across four datasets and six models."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Dynamic Latent Routing recovers globally optimal policies by composing learned sub-policies to improve low-data language model fine-tuning."}],"snapshot_sha256":"f8e4659fe691fbe54d74086337cf1ec7da3a4a2afd57db837e1e2232949e86d2"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"f1863e9f4efa9195501d90122a4c64020c051095898c9a1e196df5aca7dda042"},"paper":{"abstract_excerpt":"We investigate the temporal concatenation of sub-policies in Markov Decision Processes (MDP) with time-varying reward functions. We introduce General Dijkstra Search (GDS), and prove that globally optimal goal-reaching policies can be recovered through temporal composition of intermediate optimal sub-policies. Motivated by the \"search, select, update\" principle underlying GDS, we propose Dynamic Latent Routing (DLR), a language-model post-training method that jointly learns discrete latent codes, routing policies, and model parameters through dynamic search in a single training stage. In low-d","authors_text":"Amir Abdullah, Fangyuan Yu, Xin Su","cross_cats":["cs.AI","cs.CL"],"headline":"Dynamic Latent Routing recovers globally optimal policies by composing learned sub-policies to improve low-data language model fine-tuning.","license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T03:35:46Z","title":"Dynamic Latent Routing"},"references":{"count":70,"internal_anchors":4,"resolved_work":70,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Hindsight experience replay","work_id":"b4ff437e-eec1-45ab-9ee2-e70f0e4c45a4","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Introducing Claude Opus 4.7","work_id":"93e7fbe6-fe60-46cb-9cca-d936e25cd4c7","year":2026},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"The option-critic architecture","work_id":"eca5b255-3362-4279-8e47-baa1fa132831","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Successor features for transfer in reinforcement learning","work_id":"0c3776bc-e342-4c21-adda-1ff1b22ce73c","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Probing classifiers: Promises, shortcomings, and advances.Computational Linguistics, 48(1):207–219, 2022","work_id":"14d0b945-57f3-41b0-b491-72316d4c06b5","year":2022}],"snapshot_sha256":"de5cc2c99271703570f8a5c460fae39d8ff7b0ade229e6cc096927452a5384a8"},"source":{"id":"2605.14323","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T01:45:27.516595Z","id":"16c71107-db8a-4430-acef-4ad0a8557dd1","model_set":{"reader":"grok-4.3"},"one_line_summary":"Dynamic Latent Routing jointly learns discrete latent codes, routing policies, and model parameters via dynamic search to match or exceed supervised fine-tuning by 6.6 points on average in low-data settings across four datasets and six models.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Dynamic Latent Routing recovers globally optimal policies by composing learned sub-policies to improve low-data language model fine-tuning.","strongest_claim":"In low-data fine-tuning settings, DLR matches or outperforms supervised fine-tuning across four datasets and six models, achieving a mean gain of +6.6 percentage points, while prior discrete-latent baselines consistently underperform SFT.","weakest_assumption":"That the optimality guarantees and search principle from General Dijkstra Search in MDPs transfer effectively to the non-stationary, high-dimensional setting of language model post-training without introducing hidden biases or optimization instabilities."}},"verdict_id":"16c71107-db8a-4430-acef-4ad0a8557dd1"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:b2d68aeec0acbfc54168ac7f336c56f02451455a884e2cf8064fa56fa67df122","target":"record","created_at":"2026-05-17T23:39:09Z","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":"07849c9d1a4bf545b3e2a82e4b983491ba8c311b3d8a1b2b18df2693d350a8f2","cross_cats_sorted":["cs.AI","cs.CL"],"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T03:35:46Z","title_canon_sha256":"053ff61afba5cb9c959c212834e1d656d977b42a269996ce53447947c04bb37a"},"schema_version":"1.0","source":{"id":"2605.14323","kind":"arxiv","version":1}},"canonical_sha256":"acb659fb39d2a38fcecd7d56bb52aabbd5954034aff8d68f28aefd97cfaf27d8","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"acb659fb39d2a38fcecd7d56bb52aabbd5954034aff8d68f28aefd97cfaf27d8","first_computed_at":"2026-05-17T23:39:09.808381Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:09.808381Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Cvrvy8m5gKlxyvfdiQAUq/4Awz0PTQAY7tuGNZb094kQ2KaYjQhB3gG/UB6lOcAKC8HWceaTwgxelU7JohDoCg==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:09.808923Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14323","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:b2d68aeec0acbfc54168ac7f336c56f02451455a884e2cf8064fa56fa67df122","sha256:015b457a7ca8f04067702ce48f7ec3781377c1dfc302a45def00fccd85430c7a"],"state_sha256":"b7b8049f67549df733aefa857eabcac49aeefb8e6b6f0e73b52afd8f8ed8b6e6"}