Metraipolis › archive › dc_3d6489ba3547
BRIDGE: Bilevel Retrieval-Credit-Aware Agentic Reinforcement Learning — arXiv:2609.36505
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.36505
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T16:22:41.724Z
- SHA-256 of the archived text
b88fb4c1698e96b38ffe5b0b670acc79e162e4163d4d1ec87f27ed146a8b7db2- Stable URL of this page
https://metraipolis.com/documento/dc_3d6489ba3547
How to cite
BRIDGE: Bilevel Retrieval-Credit-Aware Agentic Reinforcement Learning — arXiv:2609.36505 — arXiv (en), https://arxiv.org/abs/2609.36505 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3d6489ba3547 (Metraipolis archive dc_3d6489ba3547, sha256 b88fb4c1698e96b38ffe5b0b670acc79e162e4163d4d1ec87f27ed146a8b7db2)
BibTeX
@misc{dc3d6489ba3547,
title = {BRIDGE: Bilevel Retrieval-Credit-Aware Agentic Reinforcement Learning — arXiv:2609.36505},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.36505},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3d6489ba3547; sha256 b88fb4c1698e96b38ffe5b0b670acc79e162e4163d4d1ec87f27ed146a8b7db2},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36505. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Agentic reinforcement learning (ARL) with verifiable rewards improves the ability of large language models (LLMs) to tackle knowledge-intensive tasks by learning to interleave search and reasoning. However, most existing ARL methods optimize only LLM-generated tokens and treat retrieved evidence as environment observations. This creates an information-credit gap: failures caused by missing or misleading evidence are attributed to the LLM policy rather than to the retriever, which motivates training the LLM and the retriever jointly. In this paper, we show that retrieval and LLM policy learning are order-sensitive: adapting the retriever before optimizing the policy yields a larger reward gain than the reverse order. To preserve this hierarchy while allowing both components to co-adapt, we formulate retrieval-augmented agentic RL as a bilevel optimization problem. To solve it efficiently, we introduce BRIDGE, a memory-efficient first-order bilevel method motivated by a loss-landscape analysis of the RL and retrieval objectives. Across seven open-domain QA benchmarks, BRIDGE achieves the highest average accuracy with both 3B and 7B backbones, improving the multi-hop average over the strongest baseline by 9.6 and 3.4 EM points, respectively. It also achieves the best averaged answer accuracy and reasoning quality across medical QA benchmarks. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG); Optimization and Control (math.OC). Pagina del lavoro: https://arxiv.org/abs/2609.36505 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-07T15:54:53.507Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).