arXiv:2605.31346v1 Announce Type: cross Abstract: Zeroth-order (black-box) optimization is applied when gradients are unavailable and objective evaluations rely on expensive simulations. In many such applications, the oracle fidelity is tunable: higher-accuracy queries reduce noise but incur higher computational costs. To capture this trade-off, we study an accuracy-aware wall-clock model where each query with fidelity $\delta$ has a cost $c(\delta)$, and we minimize the total time $T_{\mathrm{total}} = \sum_{k=1}^{N} c(\delta_k)$, subject to a target accuracy constraint. We show how the choic
This research addresses the optimization of expensive, black-box simulations, a challenge increasingly prevalent in advanced AI and scientific computing, reflective of current computational limitations.
The proposed method offers a way to significantly improve efficiency in AI training and complex system design where computational resources are a bottleneck, by optimizing the trade-off between query accuracy and computational cost.
Traditional optimization approaches often ignore the cost-accuracy trade-off of objective evaluations; this work introduces a framework to explicitly manage and minimize total 'wall-clock' time under accuracy constraints.
- · AI/ML researchers
- · High-performance computing (HPC) centers
- · Industries relying on complex simulations (e.g., aerospace, pharmaceuticals)
- · Cloud computing providers
- · Organizations with inefficient computation infrastructure
- · Methods relying solely on uniform-fidelity querying
More efficient resource utilization for complex AI model training and scientific simulations.
Accelerated discovery and development in fields heavily reliant on black-box optimization, potentially reducing R&D costs.
Enhanced accessibility to complex AI and simulation tools for smaller entities by lowering effective computational barriers.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG