{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:L47S34273GOIRAZOSTLRWHU7UC","short_pith_number":"pith:L47S3427","schema_version":"1.0","canonical_sha256":"5f3f2df35fd99c88832e94d71b1e9fa0a19abcc9ee60eb355addafcbf7b137e0","source":{"kind":"arxiv","id":"2505.09603","version":2},"attestation_state":"computed","paper":{"title":"DataMIL: Selecting Data for Robot Imitation Learning with Datamodels","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Alaa Khaddaj, Aleksander Madry, Andrew Ilyas, Logan Engstrom, Roberto Mart\\'in-Mart\\'in, Shivin Dass","submitted_at":"2025-05-14T17:55:10Z","abstract_excerpt":"Recently, the robotics community has amassed ever larger and more diverse datasets to train generalist policies. However, while these policies achieve strong mean performance across a variety of tasks, they often underperform on individual, specialized tasks and require further tuning on newly acquired task-specific data. Combining task-specific data with carefully curated subsets of large prior datasets via co-training can produce better specialized policies, but selecting data naively may actually harm downstream performance. To address this, we introduce DataMIL, a data selection framework "},"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":"2505.09603","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.RO","submitted_at":"2025-05-14T17:55:10Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"a83843d5f587166d79e5b1d79973bf1a3633d8261d890d2530407b067c44711a","abstract_canon_sha256":"2bc4d81a92d91460b4e8c6a9ccb81c2110fa3bed298afc75c8ea8191a776b982"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T01:12:45.711807Z","signature_b64":"FZ5zU85kmEXX/2yCZoxw2ZKymFGJj/ZkZ4Bfi3ouutGwwJqV4Ccc4fABFTTjez3jl5Qw4GCTdPX+ajO0efcODw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5f3f2df35fd99c88832e94d71b1e9fa0a19abcc9ee60eb355addafcbf7b137e0","last_reissued_at":"2026-06-23T01:12:45.711263Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T01:12:45.711263Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DataMIL: Selecting Data for Robot Imitation Learning with Datamodels","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.RO","authors_text":"Alaa Khaddaj, Aleksander Madry, Andrew Ilyas, Logan Engstrom, Roberto Mart\\'in-Mart\\'in, Shivin Dass","submitted_at":"2025-05-14T17:55:10Z","abstract_excerpt":"Recently, the robotics community has amassed ever larger and more diverse datasets to train generalist policies. However, while these policies achieve strong mean performance across a variety of tasks, they often underperform on individual, specialized tasks and require further tuning on newly acquired task-specific data. Combining task-specific data with carefully curated subsets of large prior datasets via co-training can produce better specialized policies, but selecting data naively may actually harm downstream performance. To address this, we introduce DataMIL, a data selection framework "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.09603","kind":"arxiv","version":2},"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/2505.09603/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":"2505.09603","created_at":"2026-06-23T01:12:45.711342+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.09603v2","created_at":"2026-06-23T01:12:45.711342+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.09603","created_at":"2026-06-23T01:12:45.711342+00:00"},{"alias_kind":"pith_short_12","alias_value":"L47S34273GOI","created_at":"2026-06-23T01:12:45.711342+00:00"},{"alias_kind":"pith_short_16","alias_value":"L47S34273GOIRAZO","created_at":"2026-06-23T01:12:45.711342+00:00"},{"alias_kind":"pith_short_8","alias_value":"L47S3427","created_at":"2026-06-23T01:12:45.711342+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.26349","citing_title":"Closing the Loop in Teleoperation: Episode-Level Data Quality Assessment and Feedback for High-Quality Demonstration Collection","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01544","citing_title":"An Efficient Metric for Data Quality Measurement in Imitation Learning","ref_index":24,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC","json":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC.json","graph_json":"https://pith.science/api/pith-number/L47S34273GOIRAZOSTLRWHU7UC/graph.json","events_json":"https://pith.science/api/pith-number/L47S34273GOIRAZOSTLRWHU7UC/events.json","paper":"https://pith.science/paper/L47S3427"},"agent_actions":{"view_html":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC","download_json":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC.json","view_paper":"https://pith.science/paper/L47S3427","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.09603&json=true","fetch_graph":"https://pith.science/api/pith-number/L47S34273GOIRAZOSTLRWHU7UC/graph.json","fetch_events":"https://pith.science/api/pith-number/L47S34273GOIRAZOSTLRWHU7UC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC/action/storage_attestation","attest_author":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC/action/author_attestation","sign_citation":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC/action/citation_signature","submit_replication":"https://pith.science/pith/L47S34273GOIRAZOSTLRWHU7UC/action/replication_record"}},"created_at":"2026-06-23T01:12:45.711342+00:00","updated_at":"2026-06-23T01:12:45.711342+00:00"}