budget allocation strategies

This work addresses the challenge of maximizing end-to-end success probability in structured agent workflows under hard constraints on budget and deadline. The authors propose Monte Carlo Combinatorial Planning (MCPP), a lightweight closed-loop planner that dynamically replans during execution in response to observations. MCPP employs a finite-horizon stochastic online allocation model with parallel sampling and leverages Monte Carlo simulation to estimate, in real time, the probability of successful task completion under the given constraints. Experimental results demonstrate that MCPP significantly outperforms strong baseline methods on the CodeFlow and ProofFlow benchmarks, consistently achieving higher task completion rates across diverse budget–deadline configurations. These findings validate MCPP’s effectiveness and robustness in resource-constrained scenarios.

This study addresses the challenge of efficiently identifying practically meaningful treatment effects under resource constraints and concurrent experimentation, where conventional resource allocation strategies—optimized to minimize mean squared error (MSE)—often prove suboptimal. The authors propose a novel framework that shifts the objective toward minimizing the worst-case Type II error (i.e., miss rate) by leveraging statistical power. They develop a variance inflation mechanism with a correction factor, tailored to scenarios where outcome standard deviations are either known or estimated from pilot data, and formulate optimization models under three distinct risk criteria. A fully data-driven Surrogate-S algorithm is introduced to implement the approach without requiring ground-truth variance information. Theoretical analysis demonstrates the potential inefficiency of MSE-oriented strategies in detection tasks, while numerical experiments show that the proposed method achieves near-optimal performance using only pilot-based variance estimates.

Existing data selection methods ignore computational budget constraints, leading to unstable performance across varying budgets—and sometimes even underperforming random selection. To address this, we propose Computation-Aware Data Selection (CADS), a budget-aware framework that treats computational budget as a core optimization variable and jointly models data selection and budget constraints via a bi-level optimization formulation. CADS introduces three key innovations: (i) a Hessian-free gradient estimator for efficient meta-gradient computation, (ii) probabilistic reparameterization of the selection policy, and (iii) an inner-loop penalty simplification strategy to accelerate convergence. Evaluated on diverse vision and language tasks, CADS achieves up to 14.42% absolute accuracy improvement over state-of-the-art baselines. To our knowledge, CADS is the first method to systematically incorporate computational budget into the data selection decision process, significantly enhancing both training efficiency and model generalization.

This work addresses the challenge of coupling slow resource provisioning with fast scheduling decisions in network resource allocation, where the former is constrained by switching costs and the latter must satisfy dynamically evolving budget constraints. To this end, we propose the first bilevel online learning framework that integrates Online Convex Optimization (OCO) with Constrained Markov Decision Processes (CMDPs). The upper level performs budget allocation via OCO with switching costs, while the lower level executes state-dependent safe scheduling based on CMDPs. A novel dual feedback mechanism propagates sensitivity information of budget multipliers across layers to enforce cross-level constraint coupling. Additionally, we introduce a budget-adaptive safe exploration strategy to handle dynamic constraints. Theoretical analysis shows that the proposed method achieves near-optimal cumulative regret while satisfying cross-level constraints with high probability, offering dual guarantees on both performance and feasibility.

High variance in policy evaluation for long-horizon reinforcement learning tasks—stemming from suboptimal behavior policies and baseline designs—severely limits sample efficiency. To address this, we propose a “doubly optimal” joint optimization framework that simultaneously optimizes both the behavior policy and the control variate baseline, achieving strict variance reduction while preserving unbiasedness. Our method integrates importance sampling, control variates, and policy gradients, and establishes a theoretical analysis grounded in the bias–variance trade-off. We prove that our estimator’s variance is strictly lower than that of all existing optimal estimators. Empirically, on multiple continuous-control benchmark tasks, our approach reduces estimation variance significantly under identical sample budgets, improves evaluation accuracy by over 40%, and achieves state-of-the-art (SOTA) performance.

Traditional scaling law estimation suffers from high computational costs due to the absence of efficient budget allocation strategies. This work proposes a novel approach that, for the first time, integrates surrogate-guided pruning into scaling law modeling by combining the Successive Halving algorithm with both parametric and non-parametric surrogate models. This integration enables proactive allocation of computational resources and efficient construction of loss-compute Pareto frontiers. The method substantially improves resource utilization efficiency, achieving relative performance gains of up to 2.84% on real datasets and 5.47% on synthetic datasets, while reducing computational costs by as much as 98.7%.

This study addresses the problem of learning optimal decision policies under dual constraints of a budget limit and a minimum coverage requirement, formulating it for the first time as a combinatorial optimization problem with a knapsack structure. Through linear programming relaxation and Lagrangian duality analysis, the authors establish that the problem exhibits a constant-order integrality gap and derive an affine threshold rule—based on shadow prices—that characterizes the optimal policy. Two efficient algorithms are proposed: the Greedy-Lagrangian algorithm achieves near-optimal performance with finite samples, while the rank-and-cut algorithm is provably near-optimal when costs are homogeneous or the coverage constraint is loose, exhibiting only minor misallocations in the challenging regime of heterogeneous costs and tight coverage constraints.

This work addresses the challenge of safely harnessing large language models (LLMs) in high-throughput experimental optimization, where direct LLM use risks unsafe exploration yet complete exclusion forfeits their optimization potential. To reconcile this trade-off, the authors propose the CARE framework, which employs a non-LLM default optimizer as the primary pathway while leveraging the LLM to generate candidate strategies. Adoption of these candidates is governed by an evidence-based intervention gating mechanism that audits proposals against publicly available evidence, ensuring decisions are auditable, controllable, and traceable. By synergistically integrating LLM-driven creativity with evidence-guided safety constraints, CARE achieves state-of-the-art performance on the Minerva/Olympus and ChemLex benchmarks, improving peak scores from 80.0 to 88.5 and from 83.9 to 92.1, respectively.

This work addresses the high cost of small-scale pilot experiments required to fit scaling laws in large-scale model training. Framing the problem as a budget-aware sequential experimental design task, the authors propose an uncertainty-aware active selection strategy that dynamically chooses the most informative experiments from a heterogeneous-cost pool for extrapolation to the target regime. By integrating sequential experimental design, uncertainty quantification, and active learning, the method achieves fitting accuracy comparable to that of exhaustive experimentation using only approximately 10% of the total training budget across diverse scaling law tasks, substantially outperforming conventional experimental design baselines.

This work addresses the inefficiency and suboptimality of manually tuned weight assignments in multi-objective reinforcement learning for industrial automation. It proposes the first integration of multi-objective Bayesian optimization with reinforcement learning, leveraging a sampling strategy based on the expected hypervolume improvement (qEHVI) to efficiently explore the Pareto front between energy consumption and control performance. Evaluated on the Quanser Aero 2 platform in a one-degree-of-freedom pitch control task, the method significantly outperforms uniform grid search by achieving superior hypervolume metrics and broader policy distribution coverage with substantially fewer evaluations. This demonstrates a sample-efficient approach to automatically discovering diverse Pareto-optimal policies without manual trade-off tuning.