{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3I54KFMZVI5FJQ32BMWMKZKTPX","short_pith_number":"pith:3I54KFMZ","schema_version":"1.0","canonical_sha256":"da3bc51599aa3a54c37a0b2cc565537dce05b1f2d12a72bc86237b08aef90286","source":{"kind":"arxiv","id":"2605.28678","version":1},"attestation_state":"computed","paper":{"title":"DREAM-R: Multimodal Speculative Reasoning with RL-Based Refined Drafting, Precise Verification, and Fully Parallel Execution","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bo Bao, Eric Sather, Sai Qian Zhang, Tianhua Xia, Vithursan Thangarasa, Xiangyang Yin, Yunhai Hu, Zining Liu","submitted_at":"2026-05-27T16:11:10Z","abstract_excerpt":"Speculative reasoning has recently been proposed as a means to accelerate reasoning-intensive generation in large multimodal models, but its effectiveness is often constrained by misalignment between speculative drafts and target-verified reasoning. In this work, we introduce DREAM-R, a framework that substantially improves the performance of speculative reasoning. At its core, DREAM-R employs Speculative Alignment Policy Optimization (SAPO), a reinforcement-learning objective that trains draft models to generate reasoning steps that are both faithful to target trajectories and concise. We fur"},"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":"2605.28678","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AI","submitted_at":"2026-05-27T16:11:10Z","cross_cats_sorted":[],"title_canon_sha256":"3083a9129c713f7db994d7619521eafe9f62f248cb07b56e3b5516337a57c4cd","abstract_canon_sha256":"595e3eb1be5f31d6ebc57f0ce096373c170d4fe63b9583522f6f8a7cdda2fda5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T02:04:59.643451Z","signature_b64":"B9sT6twtFr1S2Ie71N1Q/zh3EKFRZvPRLayS0NdSu4Id1x6BTVbDvfmzht0ZtMjd5BYZfpUCbI713OWDpao1BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"da3bc51599aa3a54c37a0b2cc565537dce05b1f2d12a72bc86237b08aef90286","last_reissued_at":"2026-05-28T02:04:59.643028Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T02:04:59.643028Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"DREAM-R: Multimodal Speculative Reasoning with RL-Based Refined Drafting, Precise Verification, and Fully Parallel Execution","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Bo Bao, Eric Sather, Sai Qian Zhang, Tianhua Xia, Vithursan Thangarasa, Xiangyang Yin, Yunhai Hu, Zining Liu","submitted_at":"2026-05-27T16:11:10Z","abstract_excerpt":"Speculative reasoning has recently been proposed as a means to accelerate reasoning-intensive generation in large multimodal models, but its effectiveness is often constrained by misalignment between speculative drafts and target-verified reasoning. In this work, we introduce DREAM-R, a framework that substantially improves the performance of speculative reasoning. At its core, DREAM-R employs Speculative Alignment Policy Optimization (SAPO), a reinforcement-learning objective that trains draft models to generate reasoning steps that are both faithful to target trajectories and concise. We fur"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28678","kind":"arxiv","version":1},"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/2605.28678/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":"2605.28678","created_at":"2026-05-28T02:04:59.643088+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28678v1","created_at":"2026-05-28T02:04:59.643088+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28678","created_at":"2026-05-28T02:04:59.643088+00:00"},{"alias_kind":"pith_short_12","alias_value":"3I54KFMZVI5F","created_at":"2026-05-28T02:04:59.643088+00:00"},{"alias_kind":"pith_short_16","alias_value":"3I54KFMZVI5FJQ32","created_at":"2026-05-28T02:04:59.643088+00:00"},{"alias_kind":"pith_short_8","alias_value":"3I54KFMZ","created_at":"2026-05-28T02:04:59.643088+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/3I54KFMZVI5FJQ32BMWMKZKTPX","json":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX.json","graph_json":"https://pith.science/api/pith-number/3I54KFMZVI5FJQ32BMWMKZKTPX/graph.json","events_json":"https://pith.science/api/pith-number/3I54KFMZVI5FJQ32BMWMKZKTPX/events.json","paper":"https://pith.science/paper/3I54KFMZ"},"agent_actions":{"view_html":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX","download_json":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX.json","view_paper":"https://pith.science/paper/3I54KFMZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28678&json=true","fetch_graph":"https://pith.science/api/pith-number/3I54KFMZVI5FJQ32BMWMKZKTPX/graph.json","fetch_events":"https://pith.science/api/pith-number/3I54KFMZVI5FJQ32BMWMKZKTPX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX/action/storage_attestation","attest_author":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX/action/author_attestation","sign_citation":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX/action/citation_signature","submit_replication":"https://pith.science/pith/3I54KFMZVI5FJQ32BMWMKZKTPX/action/replication_record"}},"created_at":"2026-05-28T02:04:59.643088+00:00","updated_at":"2026-05-28T02:04:59.643088+00:00"}