Metraipolis › archive › dc_f60503a85154
CircuitLens: Reasoning Circuits as Data Selection Signals for Reinforcement Learning with Verifiable Rewards — arXiv:2609.07183
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.07183
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T12:45:29.587Z
- SHA-256 of the archived text
8802144d57af63630cd5d4d8d292fa63c2878723c91a4ca2c0c8fac9c65c9202- Stable URL of this page
https://metraipolis.com/documento/dc_f60503a85154
How to cite
CircuitLens: Reasoning Circuits as Data Selection Signals for Reinforcement Learning with Verifiable Rewards — arXiv:2609.07183 — arXiv (en), https://arxiv.org/abs/2609.07183 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f60503a85154 (Metraipolis archive dc_f60503a85154, sha256 8802144d57af63630cd5d4d8d292fa63c2878723c91a4ca2c0c8fac9c65c9202)
BibTeX
@misc{dcf60503a85154,
title = {CircuitLens: Reasoning Circuits as Data Selection Signals for Reinforcement Learning with Verifiable Rewards — arXiv:2609.07183},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.07183},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f60503a85154; sha256 8802144d57af63630cd5d4d8d292fa63c2878723c91a4ca2c0c8fac9c65c9202},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.07183. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Reinforcement learning with verifiable rewards (RLVR) is sensitive to which problems a model trains on, yet existing selection criteria--difficulty filtering, hand-curation, reward-trajectory scoring--assess data value as an intrinsic property of problems, independent of the model that will learn from them. We introduce Circuit Reasoning Score (CRS), a selection signal derived from 46 reasoning-sensitive attention heads identified via contrastive ablation, computed in a single forward pass on the frozen base model without reward labels or rollouts. CRS runs against the intuitive hypothesis that stronger reasoning-circuit engagement produces better training data: on Qwen2.5-Math-7B, the lowest-engagement decile improves over random selection on three medium-difficulty benchmarks (GSM8K +2.0 pp, OlympiadBench +1.6 pp, Minerva +2.9 pp), while the highest-engagement decile gains less and is indistinguishable from the middle decile. The advantage has boundary conditions: on a domain-curated pool no selection method separates from the others; at 1.5B scale the useful direction differs; and the lowest-reward training condition produces the strongest downstream generalization. Within the Qwen2.5-Math settings tested, RLVR data selection appears regime-dependent rather than reducible to a static ranking of problem quality. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL). Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2609.07183 — 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 2026-10-16T12:30:25.100Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).