Reproduce “A Two-Layer Framework for Joint Online Configuration Selection and Admission Control” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Owen Shen, Haoran Xu, Yinyu Ye, Peter Glynn, Patrick Jaillet

SP-UCB-OLP learns a mixture of request-generating configurations and uses resource-price thresholds to accept or reject observed requests under cumulative budgets. Its switching-aware fluid benchmark allows mixing while relaxing per-period limits to average budgets. Theorem 8 proves it upper-bounds expected reward for feasible online policies; regret is Õ(α√KT) for fixed resource dimension under the stated stochastic assumptions.

Key insights

  • A fixed-configuration oracle is insufficient; dynamic switching between configurations can achieve higher rewards by exploiting complementary resources, necessitating a switching-aware benchmark.
  • The optimal strategy for the fluid oracle can be characterized by saddle point conditions involving a global "bid price" and a "surplus function" for each configuration.
  • Even minimal exploration is crucial in online configuration selection, as greedy policies can get "stuck" in suboptimal local optima, leading to significantly lower and unstable performance.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.