Metraipolis › archive › dc_f00d4de4fe63
Tracking the Moving Frontier: Long-Short Term Advantage Estimator — arXiv:2609.06671
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.06671
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T13:57:22.264Z
- SHA-256 of the archived text
583926f66e5e811765d226178f82fc3628af9c0f78178ba214ec1623eaa3ddcd- Stable URL of this page
https://metraipolis.com/documento/dc_f00d4de4fe63
How to cite
Tracking the Moving Frontier: Long-Short Term Advantage Estimator — arXiv:2609.06671 — arXiv (en), https://arxiv.org/abs/2609.06671 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f00d4de4fe63 (Metraipolis archive dc_f00d4de4fe63, sha256 583926f66e5e811765d226178f82fc3628af9c0f78178ba214ec1623eaa3ddcd)
BibTeX
@misc{dcf00d4de4fe63,
title = {Tracking the Moving Frontier: Long-Short Term Advantage Estimator — arXiv:2609.06671},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.06671},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f00d4de4fe63; sha256 583926f66e5e811765d226178f82fc3628af9c0f78178ba214ec1623eaa3ddcd},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.06671. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Group-based RLVR methods estimate advantages by repeatedly sampling multiple trajectories for each prompt, making long-horizon agent training expensive and discarding useful experience accumulated across iterations. We ask whether historical experience can replace these repeated within-iteration comparisons without directly optimizing on stale trajectories. We introduce Long-Short Term Advantage Estimator (LSTAE), a single-stream RL algorithm that uses history for advantage estimation while updating the policy only with the current rollout. LSTAE maintains a persistent tracker for each task anchor. At the trajectory level (long term), a drift-aware historical baseline tracks the anchor's moving success frontier and measures the relative contribution of each new trajectory. At the step level (short term), a recent state-experience buffer exploits recurrent states to estimate localized action advantages. This two-timescale design converts accumulated experience into multi-granular credit signals, requiring only one rollout per anchor. Across agentic and mathematical reasoning benchmarks, LSTAE matches or improves upon strong group-based baselines while substantially reducing rollout cost. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.LG). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.06671 — 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 2026-10-16T13:39:42.906Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).