Title: Improvise, Adapt, Overcome: An On-The-Fly Multifidelity Algorithm for Efficient Machine Learning

Original: arXiv:2606.02662v1 Announce Type: cross Abstract: Machine learning has accelerated quantum chemistry but is hindered by the prohibitive cost of generating high fidelity training data. Multifidelity machine learning (MFML) mitigates this overhead by systematically combining abundant low fidelity data with sparse high fidelity data. In spite of its success, standard MFML schemes rely on pre-defined scaling factors to determine sparse data ratio across fidelities, often generating redundant multifidelity data resulting in a loss of efficiency. Here, we introduce an adaptive on-the-fly multifidelity framework for machine learning that autonomously determines training dataset composition. By dynamically querying training samples at each fidelity, the algorithm saturates model accuracy at lower fidelities before moving up to more expensive reference calculations. We benchmark the novel adaptive-MFML across diverse chemical properties including the computational chemistry gold standard coupled cluster energies, and the more chemically challenging excitation energies. In our numerical experiments we show that our adaptive algorithm reduces data generation costs by up to a factor of 30 compared to single fidelity methods and improves upon standard MFML by up to a factor of 5. The mitigation of data redundancy establishes a high-accuracy low-cost pathway for sustainable cost-aware machine learning in quantum chemistry.

Rewrite: Machine learning has significantly advanced the field of quantum chemistry, yet its progress is often constrained by the exorbitant expense of producing high-fidelity training datasets. Multifidelity machine learning (MFML) addresses this bottleneck by strategically integrating plentiful low-fidelity data with limited high-fidelity samples. Despite its effectiveness, conventional MFML approaches depend on fixed scaling factors to allocate data ratios across different fidelity levels, a practice that frequently leads to redundant data and diminished efficiency. To address this, we present a novel adaptive, on-the-fly multifidelity framework that autonomously curates the composition of the training dataset. This algorithm dynamically solicits training samples at each fidelity level, ensuring that model accuracy is maximized at lower-cost stages before progressing to more resource-intensive reference calculations. We evaluated the performance of our adaptive-MFML method on a variety of chemical properties, ranging from the computational chemistry benchmark of coupled cluster energies to the more complex task of predicting excitation energies. Our numerical results demonstrate that the adaptive approach cuts data generation expenses by as much as 30 times relative to single-fidelity techniques and outperforms traditional MFML methods by a factor of up to 5. By eliminating data redundancy, this work paves the way for a sustainable, cost-effective, and high-accuracy approach to machine learning in quantum chemistry.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC