Distributionally Robust Agentic RL for Web Decision-Making under Multi-Cost Budgets and Failure Risk
Global Media and Social Sciences Research Journal · 15.03.2026 · 7 (1) · 250-256
DOI: 10.71465/gmssrj170 ↗Kurz gefasst
Fachartikel in der Zeitschrift Global Media and Social Sciences Research Journal, online erschienen am 15.03.2026. Verfasst von James A. Anderson und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Web environments are non-stationary: DOM layouts, page latency, and API pricing can shift, causing brittle policies to violate budgets or trigger failures. We model web interaction as a constrained MDP with a multi-dimensional cost vector and a failure penalty, and propose a distributionally robust learning objective that optimizes worst-case expected return over an uncertainty set of transition dynamics and cost distributions. The method, DR-ARC, combines (i) Wasserstein-robust value estimation for cost-aware Q-functions, (ii) a budget-feasible policy update using robust advantages, and (iii) a failure-risk envelope that penalizes actions with high sensitivity to dynamics shift (estimated by local Lipschitz bounds of Q w.r.t. state embeddings). Experiments are recommended on a multi-site benchmark with 1,000–2,000 tasks and site-shift splits (train on 60% sites, test on unseen 40%), including induced latency spikes and UI perturbations. Metrics include success rate, mean/variance of each cost dimension, failure rate, and robust constraint violation under perturbed test distributions. DR-ARC aims to reduce budget overruns and tail failures under realistic distribution shifts.
Zitieren
APA
Anderson, J. A., Miller, R. L., & Johnson, D. K. (2026). Distributionally Robust Agentic RL for Web Decision-Making under Multi-Cost Budgets and Failure Risk. Global Media and Social Sciences Research Journal, 7(1), 250-256. https://doi.org/10.71465/gmssrj170
Vancouver
Anderson JA, Miller RL, Johnson DK. Distributionally Robust Agentic RL for Web Decision-Making under Multi-Cost Budgets and Failure Risk. Global Media and Social Sciences Research Journal. 2026;7(1):250-256. doi:10.71465/gmssrj170
BibTeX
@article{Anderson2026,
author = {Anderson, James A. and Miller, Rebecca L. and Johnson, David K.},
title = {Distributionally Robust Agentic RL for Web Decision-Making under Multi-Cost Budgets and Failure Risk},
journal = {Global Media and Social Sciences Research Journal},
year = {2026},
volume = {7},
number = {1},
pages = {250--256},
doi = {10.71465/gmssrj170}
}
EUVIMED