{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Q5I5NJUIQQZO2DDBCP7YGPRICP","short_pith_number":"pith:Q5I5NJUI","schema_version":"1.0","canonical_sha256":"8751d6a6888432ed0c6113ff833e2813f2298b74ce62a8a5ea6a3a9abf5f0472","source":{"kind":"arxiv","id":"2501.14513","version":2},"attestation_state":"computed","paper":{"title":"ABPT: Amended Backpropagation through Time with Partially Differentiable Rewards","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.RO","authors_text":"Danping Zou, Fangyu Sun, Fanxing Li, Tianbao Zhang","submitted_at":"2025-01-24T14:18:22Z","abstract_excerpt":"Quadrotor control policies can be trained with high performance using the exact gradients of the rewards to directly optimize policy parameters via backpropagation-through-time (BPTT). However, designing a fully differentiable reward architecture is often challenging. Partially differentiable rewards will result in biased gradient propagation that degrades training performance. To overcome this limitation, we propose Amended Backpropagation-through-Time (ABPT), a novel approach that mitigates gradient bias while preserving the training efficiency of BPTT. ABPT combines 0-step and N-step return"},"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":"2501.14513","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.RO","submitted_at":"2025-01-24T14:18:22Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"49adc65eb85c76f10a92f0df4718e59b0de71dfacd2dc7c8fad6d2c403bdb044","abstract_canon_sha256":"22ebb83b2149ef62ac9f0eb2929b173ca4a850112e655c1f20c090475f1c2eaf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:06:27.223897Z","signature_b64":"fNqOdhP8n1oMnKDfcbQlaQZOdZwxFINAso6ZFJqZ/5UrlV7kWs/Yec3QOfDXLhki0co8EWSQfz9Zqmp5aL6eBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8751d6a6888432ed0c6113ff833e2813f2298b74ce62a8a5ea6a3a9abf5f0472","last_reissued_at":"2026-07-05T11:06:27.223414Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:06:27.223414Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ABPT: Amended Backpropagation through Time with Partially Differentiable Rewards","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.RO","authors_text":"Danping Zou, Fangyu Sun, Fanxing Li, Tianbao Zhang","submitted_at":"2025-01-24T14:18:22Z","abstract_excerpt":"Quadrotor control policies can be trained with high performance using the exact gradients of the rewards to directly optimize policy parameters via backpropagation-through-time (BPTT). However, designing a fully differentiable reward architecture is often challenging. Partially differentiable rewards will result in biased gradient propagation that degrades training performance. To overcome this limitation, we propose Amended Backpropagation-through-Time (ABPT), a novel approach that mitigates gradient bias while preserving the training efficiency of BPTT. ABPT combines 0-step and N-step return"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.14513","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/2501.14513/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":"2501.14513","created_at":"2026-07-05T11:06:27.223468+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.14513v2","created_at":"2026-07-05T11:06:27.223468+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.14513","created_at":"2026-07-05T11:06:27.223468+00:00"},{"alias_kind":"pith_short_12","alias_value":"Q5I5NJUIQQZO","created_at":"2026-07-05T11:06:27.223468+00:00"},{"alias_kind":"pith_short_16","alias_value":"Q5I5NJUIQQZO2DDB","created_at":"2026-07-05T11:06:27.223468+00:00"},{"alias_kind":"pith_short_8","alias_value":"Q5I5NJUI","created_at":"2026-07-05T11:06:27.223468+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.10548","citing_title":"Simple but Stable, Fast and Safe: Achieve End-to-end Control by High-Fidelity Differentiable Simulation","ref_index":42,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP","json":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP.json","graph_json":"https://pith.science/api/pith-number/Q5I5NJUIQQZO2DDBCP7YGPRICP/graph.json","events_json":"https://pith.science/api/pith-number/Q5I5NJUIQQZO2DDBCP7YGPRICP/events.json","paper":"https://pith.science/paper/Q5I5NJUI"},"agent_actions":{"view_html":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP","download_json":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP.json","view_paper":"https://pith.science/paper/Q5I5NJUI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.14513&json=true","fetch_graph":"https://pith.science/api/pith-number/Q5I5NJUIQQZO2DDBCP7YGPRICP/graph.json","fetch_events":"https://pith.science/api/pith-number/Q5I5NJUIQQZO2DDBCP7YGPRICP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP/action/storage_attestation","attest_author":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP/action/author_attestation","sign_citation":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP/action/citation_signature","submit_replication":"https://pith.science/pith/Q5I5NJUIQQZO2DDBCP7YGPRICP/action/replication_record"}},"created_at":"2026-07-05T11:06:27.223468+00:00","updated_at":"2026-07-05T11:06:27.223468+00:00"}