Pooling and Drift in Delayed Bandits
Abstract
A system often has to act long before it learns whether the act worked: a recommender sees a click in seconds and a purchase in days. With K actions and a delay of d rounds, the best rate known for this setting is Õ(√((K+d)T)) over T rounds, so a longer menu is always more expensive to learn from. It need not be: if the outcome depends on the action only through the state it produced, then one late outcome informs every action that could have produced the observed state, and the price is set by how many genuinely different states the actions produce rather than by how many actions there are. We measure this using an effective dimension vt between $1$ and the number of states, and prove Õ(√((d+1)Vlog K)) for a rotating algorithm and Õ(√(V−)+√(dT)) for the single-copy algorithm used in practice, for any budget fixed in advance; merging similar states lowers the price further, at an explicit bias. Even when given the exact losses from d rounds ago, no algorithm escapes Ω(√(dEmin{1+log J,T/d})), where J counts the drifting directions and E bounds how far losses move while the learner waits. On generated data, the state channel cuts regret by up to 79 percent against action-level weighting and, on the funnel family, by 32 to 68 percent against a tuned minimax-optimal method.
Related
- Design and Analysis Strategies for Pooling in High Throughput Screening: Application to the Search for a New Anti‐Microbial
- Transversal Pooling Neural Networks
- The Geometry of Informativeness: A Bregman-Centroid Diagnostic for Partial Pooling in Multisite Trials
- Quantifying Portfolio Demutualization: A Benchmark-Relative Pooling--Profiling Scale
- Target-Constrained Pooling for Trajectory Extrapolation
BibTeX
@misc{baghi2026pooling,
title = {{Pooling and Drift in Delayed Bandits}},
author = {Melika Baghi},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.01761},
eprint = {2609.01761},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.01761},
}
RIS
TY - UNPB
TI - Pooling and Drift in Delayed Bandits
AU - Melika Baghi
PY - 2026
DA - 2026/09/06
PB - arXiv
DO - 10.48550/arxiv.2609.01761
UR - https://arxiv.org/abs/2609.01761
AB - A system often has to act long before it learns whether the act worked: a recommender sees a click in seconds and a purchase in days. With K actions and a delay of d rounds, the best rate known for this setting is Õ(√((K+d)T)) over T rounds, so a longer menu is always more expensive to learn from. It need not be: if the outcome depends on the action only through the state it produced, then one late outcome informs every action that could have produced the observed state, and the price is set by how many genuinely different states the actions produce rather than by how many actions there are. We measure this using an effective dimension v_t between $1$ and the number of states, and prove Õ(√((d+1)Vlog K)) for a rotating algorithm and Õ(√(V^−)+√(dT)) for the single-copy algorithm used in practice, for any budget fixed in advance; merging similar states lowers the price further, at an explicit bias. Even when given the exact losses from d rounds ago, no algorithm escapes Ω(√(dEmin{1+log J,T/d})), where J counts the drifting directions and E bounds how far losses move while the learner waits. On generated data, the state channel cuts regret by up to 79 percent against action-level weighting and, on the funnel family, by 32 to 68 percent against a tuned minimax-optimal method.
ER -