{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:V2TOX4R4MOSZKMXIDCPDL6CRLQ","short_pith_number":"pith:V2TOX4R4","schema_version":"1.0","canonical_sha256":"aea6ebf23c63a59532e8189e35f8515c0cfbbef005e5f4b9cc087d0227ff4bed","source":{"kind":"arxiv","id":"2507.12612","version":3},"attestation_state":"computed","paper":{"title":"Learning What Matters: Probabilistic Task Selection via Mutual Information for Model Finetuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ganesh Ramakrishnan, Krishnateja Killamsetty, Nikhil Shivakumar Nayak, Parth Pratim Chatterjee, Prateek Chanda, Saral Sureka","submitted_at":"2025-07-16T20:14:55Z","abstract_excerpt":"Supervised fine-tuning performance for large language models depends strongly on how training budget is distributed across a heterogeneous set of tasks. In practice, mixtures are often fixed using simple heuristics (e.g., uniform or size-proportional sampling) that ignore task interactions, which can hurt transfer and waste budget on redundant sources. We introduce TaskPGM, a framework for learning continuous task mixtures via an energy-based model over tasks. Tasks form the nodes of a Markov random field: unary potentials capture per-task utility, and pairwise potentials encode inter-task rel"},"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":"2507.12612","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-16T20:14:55Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"617c2926ad5d67d6115e561e479e6701a0845161df79cffa01509c9d3e84265a","abstract_canon_sha256":"6077328156e20fa8f57c9e7f3b28babbd24d00491a0819119b9cad73dd3ac555"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:15:14.859652Z","signature_b64":"GlSbO89V6rMcXy52bVT5LfvgQcOo6DX5INyDdXg1x6r/0oLPHMMTrygM+q3HQk6k0yp7+hAstrPTu0fOUUH+Cg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"aea6ebf23c63a59532e8189e35f8515c0cfbbef005e5f4b9cc087d0227ff4bed","last_reissued_at":"2026-06-05T01:15:14.858885Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:15:14.858885Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learning What Matters: Probabilistic Task Selection via Mutual Information for Model Finetuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ganesh Ramakrishnan, Krishnateja Killamsetty, Nikhil Shivakumar Nayak, Parth Pratim Chatterjee, Prateek Chanda, Saral Sureka","submitted_at":"2025-07-16T20:14:55Z","abstract_excerpt":"Supervised fine-tuning performance for large language models depends strongly on how training budget is distributed across a heterogeneous set of tasks. In practice, mixtures are often fixed using simple heuristics (e.g., uniform or size-proportional sampling) that ignore task interactions, which can hurt transfer and waste budget on redundant sources. We introduce TaskPGM, a framework for learning continuous task mixtures via an energy-based model over tasks. Tasks form the nodes of a Markov random field: unary potentials capture per-task utility, and pairwise potentials encode inter-task rel"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.12612","kind":"arxiv","version":3},"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/2507.12612/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":"2507.12612","created_at":"2026-06-05T01:15:14.858995+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.12612v3","created_at":"2026-06-05T01:15:14.858995+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.12612","created_at":"2026-06-05T01:15:14.858995+00:00"},{"alias_kind":"pith_short_12","alias_value":"V2TOX4R4MOSZ","created_at":"2026-06-05T01:15:14.858995+00:00"},{"alias_kind":"pith_short_16","alias_value":"V2TOX4R4MOSZKMXI","created_at":"2026-06-05T01:15:14.858995+00:00"},{"alias_kind":"pith_short_8","alias_value":"V2TOX4R4","created_at":"2026-06-05T01:15:14.858995+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/V2TOX4R4MOSZKMXIDCPDL6CRLQ","json":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ.json","graph_json":"https://pith.science/api/pith-number/V2TOX4R4MOSZKMXIDCPDL6CRLQ/graph.json","events_json":"https://pith.science/api/pith-number/V2TOX4R4MOSZKMXIDCPDL6CRLQ/events.json","paper":"https://pith.science/paper/V2TOX4R4"},"agent_actions":{"view_html":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ","download_json":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ.json","view_paper":"https://pith.science/paper/V2TOX4R4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.12612&json=true","fetch_graph":"https://pith.science/api/pith-number/V2TOX4R4MOSZKMXIDCPDL6CRLQ/graph.json","fetch_events":"https://pith.science/api/pith-number/V2TOX4R4MOSZKMXIDCPDL6CRLQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ/action/storage_attestation","attest_author":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ/action/author_attestation","sign_citation":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ/action/citation_signature","submit_replication":"https://pith.science/pith/V2TOX4R4MOSZKMXIDCPDL6CRLQ/action/replication_record"}},"created_at":"2026-06-05T01:15:14.858995+00:00","updated_at":"2026-06-05T01:15:14.858995+00:00"}