{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:OG7IS5F32YRFHAAHZUPDK35DSA","short_pith_number":"pith:OG7IS5F3","schema_version":"1.0","canonical_sha256":"71be8974bbd622538007cd1e356fa39032960fec5cd844035b7efae252557132","source":{"kind":"arxiv","id":"2605.03964","version":2},"attestation_state":"computed","paper":{"title":"Pretrained Model Representations as Acquisition Signals for Active Learning of MLIPs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Pretrained MLIP latent spaces supply acquisition signals that reduce active learning data needs for reactive chemistry.","cross_cats":["physics.chem-ph"],"primary_cat":"cs.LG","authors_text":"Eszter Varga-Umbrich, Jules Tilly, Olivier Peltre, Paul Duckworth, Shikha Surana, Zachary Weller-Davies","submitted_at":"2026-05-05T16:48:23Z","abstract_excerpt":"Training machine learning interatomic potentials (MLIPs) for reactive chemistry is often bottlenecked by the high cost of quantum chemical labels and the scarcity of transition state configurations in candidate pools. Active learning (AL) can mitigate these costs, but its effectiveness hinges on the acquisition rule. We investigate whether the latent space of a pretrained MLIP already contains the information necessary for effective acquisition, eliminating the need for auxiliary uncertainty heads, Bayesian training and fine-tuning, or committee ensembles. We introduce two acquisition signals "},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.03964","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-05T16:48:23Z","cross_cats_sorted":["physics.chem-ph"],"title_canon_sha256":"59fad99b6b837a93829fb4fb40e66fe22db617b0c746ed269a7acde81b7499ce","abstract_canon_sha256":"c503f41439c75f83afd823e0a93b25e131e0f850d3fb8e02b1351b595578f1f5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:01:42.689846Z","signature_b64":"YkYl2BvHBrOAqiRhqDQhAF7aT2XB4a4OMPv4pbplR4VeTS8hrlIsiWuPYCSnQqgxTbuzHkRP/IM3NT78mbdKBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"71be8974bbd622538007cd1e356fa39032960fec5cd844035b7efae252557132","last_reissued_at":"2026-05-20T00:01:42.689244Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:01:42.689244Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pretrained Model Representations as Acquisition Signals for Active Learning of MLIPs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Pretrained MLIP latent spaces supply acquisition signals that reduce active learning data needs for reactive chemistry.","cross_cats":["physics.chem-ph"],"primary_cat":"cs.LG","authors_text":"Eszter Varga-Umbrich, Jules Tilly, Olivier Peltre, Paul Duckworth, Shikha Surana, Zachary Weller-Davies","submitted_at":"2026-05-05T16:48:23Z","abstract_excerpt":"Training machine learning interatomic potentials (MLIPs) for reactive chemistry is often bottlenecked by the high cost of quantum chemical labels and the scarcity of transition state configurations in candidate pools. Active learning (AL) can mitigate these costs, but its effectiveness hinges on the acquisition rule. We investigate whether the latent space of a pretrained MLIP already contains the information necessary for effective acquisition, eliminating the need for auxiliary uncertainty heads, Bayesian training and fine-tuning, or committee ensembles. We introduce two acquisition signals "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"On reactive-chemistry benchmarks, both kernels consistently outperform fixed-descriptor baselines, committee disagreement, and random acquisition, reducing the data required to reach performance targets by an average of 38% for energy error and 28% for force error.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The latent space of a pretrained MLIP already contains the information necessary for effective acquisition, eliminating the need for auxiliary uncertainty heads, Bayesian training and fine-tuning, or committee ensembles.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Kernels from pretrained MLIP latent spaces outperform standard acquisition methods in active learning for reactive chemistry, reducing required labels by 38% for energy error and 28% for force error.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Pretrained MLIP latent spaces supply acquisition signals that reduce active learning data needs for reactive chemistry.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ca4e78facee4840d22b64e74735f6b6c5326f4fe56f28f894cba5323e73a7429"},"source":{"id":"2605.03964","kind":"arxiv","version":2},"verdict":{"id":"c89dc312-afa0-424a-9d9b-96583823825e","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T16:40:56.767844Z","strongest_claim":"On reactive-chemistry benchmarks, both kernels consistently outperform fixed-descriptor baselines, committee disagreement, and random acquisition, reducing the data required to reach performance targets by an average of 38% for energy error and 28% for force error.","one_line_summary":"Kernels from pretrained MLIP latent spaces outperform standard acquisition methods in active learning for reactive chemistry, reducing required labels by 38% for energy error and 28% for force error.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The latent space of a pretrained MLIP already contains the information necessary for effective acquisition, eliminating the need for auxiliary uncertainty heads, Bayesian training and fine-tuning, or committee ensembles.","pith_extraction_headline":"Pretrained MLIP latent spaces supply acquisition signals that reduce active learning data needs for reactive chemistry."},"integrity":{"clean":false,"summary":{"advisory":0,"critical":1,"by_detector":{"doi_compliance":{"total":1,"advisory":0,"critical":1,"informational":0}},"informational":0},"endpoint":"/pith/2605.03964/integrity.json","findings":[{"note":"Identifier '10.1063/5.0288994' is syntactically valid but the DOI registry (doi.org) returned 404, and Crossref / OpenAlex / internal corpus also have no record. The cited work could not be located through any authoritative source.","detector":"doi_compliance","severity":"critical","ref_index":5,"audited_at":"2026-05-19T14:53:37.116701Z","detected_doi":"10.1063/5.0288994","finding_type":"unresolvable_identifier","verdict_class":"cross_source","detected_arxiv_id":null}],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-20T00:01:21.378409Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T14:53:37.116701Z","status":"completed","version":"1.0.0","findings_count":1}],"snapshot_sha256":"acad7e7452bf5df11713b32565d5bf3787c034e6eb9fc7f5d1fd11eba7d655a6"},"references":{"count":109,"sample":[{"doi":"10.1103/physrevb.87.184115","year":2013,"title":"On representing chemical environments","work_id":"e9773ddd-1a8c-4ce0-81d4-f4cff1f037ff","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"A foundation model for atomistic materials chemistry","work_id":"29bfbd6f-762e-4448-8f07-4f24bf553e51","ref_index":4,"cited_arxiv_id":"2401.00096","is_internal_anchor":true},{"doi":"","year":2023,"title":"doi:10.48550/arXiv.2206.07697 , title =","work_id":"c256ce9a-079f-49e3-ac8e-5c51a3d9dcf6","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1063/5.0288994","year":2025,"title":"L., Marsalek, O., and Schran, C","work_id":"e35c14d7-5137-44a1-9898-474a0180a307","ref_index":6,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1103/physrevlett.98.146401","year":2007,"title":"Generalized Neural-Network Representation of High-Dimensional Potential-Energy Surfaces","work_id":"1ef6c602-4928-4db2-a422-c82715f761db","ref_index":7,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":109,"snapshot_sha256":"b04be80bbaca7494b048a27d434b9082f20f7671f9789869b8bfef8f9d466bcb","internal_anchors":1},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e3cc9a9951435587e8e35c7730a3bb9abb622232ac3a712a3a0196270ff5232c"},"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":"2605.03964","created_at":"2026-05-20T00:01:42.689355+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.03964v2","created_at":"2026-05-20T00:01:42.689355+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.03964","created_at":"2026-05-20T00:01:42.689355+00:00"},{"alias_kind":"pith_short_12","alias_value":"OG7IS5F32YRF","created_at":"2026-05-20T00:01:42.689355+00:00"},{"alias_kind":"pith_short_16","alias_value":"OG7IS5F32YRFHAAH","created_at":"2026-05-20T00:01:42.689355+00:00"},{"alias_kind":"pith_short_8","alias_value":"OG7IS5F3","created_at":"2026-05-20T00:01:42.689355+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.13788","citing_title":"Force-Aware Neural Tangent Kernels for Scalable and Robust Active Learning of MLIPs","ref_index":52,"is_internal_anchor":true},{"citing_arxiv_id":"2605.13788","citing_title":"Force-Aware Neural Tangent Kernels for Scalable and Robust Active Learning of MLIPs","ref_index":52,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA","json":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA.json","graph_json":"https://pith.science/api/pith-number/OG7IS5F32YRFHAAHZUPDK35DSA/graph.json","events_json":"https://pith.science/api/pith-number/OG7IS5F32YRFHAAHZUPDK35DSA/events.json","paper":"https://pith.science/paper/OG7IS5F3"},"agent_actions":{"view_html":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA","download_json":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA.json","view_paper":"https://pith.science/paper/OG7IS5F3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.03964&json=true","fetch_graph":"https://pith.science/api/pith-number/OG7IS5F32YRFHAAHZUPDK35DSA/graph.json","fetch_events":"https://pith.science/api/pith-number/OG7IS5F32YRFHAAHZUPDK35DSA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA/action/storage_attestation","attest_author":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA/action/author_attestation","sign_citation":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA/action/citation_signature","submit_replication":"https://pith.science/pith/OG7IS5F32YRFHAAHZUPDK35DSA/action/replication_record"}},"created_at":"2026-05-20T00:01:42.689355+00:00","updated_at":"2026-05-20T00:01:42.689355+00:00"}