{"paper":{"title":"R2R2: Robust Representation for Intensive Experience Reuse via Redundancy Reduction in Self-Predictive Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A non-centered objective in self-predictive learning resolves zero-centering conflicts to stabilize representations under intensive experience reuse.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Donghyeok Lee, Jinsik Kim, Sanghyeob Song, Sungroh Yoon","submitted_at":"2026-05-13T18:38:32Z","abstract_excerpt":"For reinforcement learning in data-scarce domains like real-world robotics, intensive data reuse enhances efficiency but induces overfitting. While prior works focus on critic bias, representation-level instability in Self-Predictive Learning (SPL) under high Update-to-Data (UTD) regimes remains underexplored. To bridge this gap, we propose Robust Representation via Redundancy Reduction (R2R2), a regularization method within SPL. We theoretically identify that standard zero-centering conflicts with SPL's spectral properties and design a non-centered objective accordingly. We verify R2R2 on SPL"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"At a UTD ratio of 20, R2R2 improves TD7 by ~22% and provides additional gains on top of SimbaV2-SPL, which itself establishes a new state-of-the-art.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the identified conflict between standard zero-centering and SPL spectral properties is the primary driver of representation instability, and that the proposed non-centered objective directly causes the observed performance gains rather than other unstated experimental factors.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"R2R2 introduces a non-centered regularization objective for SPL that addresses conflicts with spectral properties, leading to better performance on continuous control tasks at high UTD ratios.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A non-centered objective in self-predictive learning resolves zero-centering conflicts to stabilize representations under intensive experience reuse.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"aefa671ad21997ecae1d5539df86117cefa6af205f77dfabfb5f5245541e36fa"},"source":{"id":"2605.14026","kind":"arxiv","version":1},"verdict":{"id":"ab62cd44-d4b5-4592-823c-42989dfcc111","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:36:09.905668Z","strongest_claim":"At a UTD ratio of 20, R2R2 improves TD7 by ~22% and provides additional gains on top of SimbaV2-SPL, which itself establishes a new state-of-the-art.","one_line_summary":"R2R2 introduces a non-centered regularization objective for SPL that addresses conflicts with spectral properties, leading to better performance on continuous control tasks at high UTD ratios.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the identified conflict between standard zero-centering and SPL spectral properties is the primary driver of representation instability, and that the proposed non-centered objective directly causes the observed performance gains rather than other unstated experimental factors.","pith_extraction_headline":"A non-centered objective in self-predictive learning resolves zero-centering conflicts to stabilize representations under intensive experience reuse."},"references":{"count":42,"sample":[{"doi":"","year":2016,"title":"Lillicrap and Jonathan J","work_id":"b2f60bdb-e387-4502-807f-af283c830418","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"9th International Conference on Learning Representations,","work_id":"34df5ed2-87c9-4393-82d1-808761a336ba","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor , booktitle =","work_id":"8231c64b-d5a3-4094-b395-b3c5345b3493","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Soft Actor-Critic Algorithms and Applications","work_id":"bb49c9fb-03b2-4226-9edb-50186b8193e4","ref_index":4,"cited_arxiv_id":"1812.05905","is_internal_anchor":true},{"doi":"","year":2025,"title":"Forty-second International Conference on Machine Learning,","work_id":"21e4bb5e-7b70-4c60-a752-713946f24abf","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":42,"snapshot_sha256":"879998858db84e2a47281d2dd79b10fe33f0a2bf720aa0e8a3e855e154720c5b","internal_anchors":5},"formal_canon":{"evidence_count":2,"snapshot_sha256":"05a0143671e977dce7717ccb14ecaad60de682e70912978fa40f6364ccf297e5"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}