This work benchmarks eleven transfer-learning surrogates against four GP methods under an identical selection rule, fidelity budget, and model size, across nine tasks spanning synthetic functions to real chemistry and materials problems, where transfer-learning surrogates reach substantially better solutions using far less computation.
Abstract
Self-driving laboratories increasingly rely on multi-fidelity Bayesian optimization (MFBO) to balance cheap, approximate evaluations against scarce, expensive ones, with a predictive surrogate at its core. Gaussian processes (GPs) are the default choice, but they scale poorly as data accumulate and assume a smooth landscape that molecular and materials search spaces routinely violate. Transfer learning offers an alternative suited to this regime: it learns a representation from abundant cheap data and adapts it to sparse expensive data. Despite its use in property prediction, transfer learning has not been tested as the engine of a closed-loop optimization. Here we benchmark eleven transfer-learning surrogates against four GP methods under an identical selection rule, fidelity budget, and model size, across nine tasks spanning synthetic functions to real chemistry and materials problems. GPs win on smooth, low-dimensional functions but perform worst on molecular and materials problems, where transfer-learning surrogates reach substantially better solutions using far less computation. Because acquisition policy is held fixed across surrogates, this advantage is attributable to the surrogate itself. Uncertainty-driven exploration is not reliably beneficial, and calibration does not predict optimization performance, so greedy exploitation of the transfer-learned mean is the more robust default. Transfer learning is therefore the surrogate of choice for molecular and materials MFBO.
Black-box optimization is a ubiquitous problem in science and engineering, often dealing with expensive objective functions with cheaper lower-fidelity proxies available. Multi-fidelity Bayesian optimization (MF-BO) is a principled approach to this problem, leveraging correlations across different fidelities when querying the objective. However, for many important MF-BO tasks, the true highest-fidelity function is prohibitively expensive to be part of the optimization loop. Nevertheless, practitioners often have gold standard data (observations of the highest-fidelity function) obtained from previous experiments that might provide information for the current task. For instance, in molecular optimization, chemists often pick the top-$k$ candidate molecules using various computer simulations, and later reveal their true objective function values. In this work, we demonstrate the suboptimality of standard MF-BO algorithms in the real-world scenarios above, even under ideal assumptions. Next, we mitigate this problem by incorporating historical high-fidelity data accompanied by task descriptors---which can be explicitly given or extracted from unstructured metadata. We demonstrate the effectiveness of our methods on synthetic functions, as well as real-world problems in chemistry and hyperparameter optimization.
G. Sutter, Hao Wang, Luis A. Ricardez-Sandoval et al.· 0 citations
Bayesian Optimization (BO) is widely adopted for data-efficient optimization in scientific and engineering applications, yet its computational cost is rarely evaluated alongside optimization performance. Here we present a systematic, compute-aware study of BO that evaluates surrogate models along two axes: optimization quality and computational frugality. Across eight benchmark functions and nine real-world datasets spanning materials science, mechanics, robotics, chemistry, and machine learning, we benchmark four surrogate models: Gaussian Processes, Random Forests, NGBoost, and Bayesian Adaptive Spline Surfaces. We show that Gaussian Process-based BO consistently incurs the highest time and memory overhead without delivering superior optimization or sample efficiency. In contrast, scalable alternatives achieve equal or better performance at a fraction of the computational cost. Motivated by these findings, we introduce a surrogate-recommendation framework that predicts the most suitable BO surrogate from inexpensive dataset characteristics. Together, these results establish FruBO as a reproducible, compute-aware baseline for Bayesian Optimization and provide practical guidance for surrogate selection under limited computational and experimental budgets.
P. Krokidas, C. Rekatsinas, Vassilis Sioros et al.· 1 citation
Machine learning offers a promising avenue to accelerate physical simulations by replacing computationally expensive traditional Partial Differential Equation (PDE) solvers with fast, differentiable surrogate models. However, standard auto-regressive ML emulators often suffer from error accumulation over long horizons and struggle to capture the stochasticity of complex physical systems. In this paper, we propose Kastor, a comprehensive methodology to adapt a deterministic physics foundation model into a highly efficient and accurate generative surrogate. First, we introduce a two-stage inference scheme that combines a large-stride causal auto-regressive model with a non-causal temporal super-resolution network, significantly reducing error accumulation while minimizing computational cost. Second, we present Mean prediction regularization (MPR), a novel training objective that constrains the generative model to predict the deterministic distribution mean under null noise conditioning. This regularization dramatically improves the performance and stability of both Functional Generative Networks (FGN) and diffusion-based emulators. Finally, we demonstrate that incorporating spatial gradient matching improves the accuracy and physical fidelity of the simulations as measured by power spectrum density. Extensive evaluations on diverse simulation datasets of the benchmark The Well show that with these components, our model outperforms competing methods in forecasting accuracy, spectral consistency, and computational efficiency. Our model achieves a 42.9% average reduction in forecasting compared to our reference based on the Walrus finetuning methodology, and outperforms Walrus for 8 out of 10 datasets on variance-normalized RMSE (VRMSE).
Guillaume Couairon, Alexis Jacq, Yu-Han Wu et al.· 0 citations
Hyperparameter selection remains a key challenge in Bayesian optimization (BO) and Bayesian active learning (AL), as model misspecification can lead to suboptimal performance, while more accurate fully Bayesian treatments typically rely on computationally expensive MCMC sampling. This paper proposes a unified framework, KENDO (Kernel ENsemble Disagreement-aware Operator), that integrates Ensemble Gaussian Processes (EGP) with disagreement-aware acquisition strategies. The central idea is to replace hyperparameter sampling with a kernel ensemble and adaptive Bayesian weighting, combined with disagreement-aware acquisition strategies. Within this unified framework, we instantiate KENDO-BO for BO and KENDO-AL for Bayesian AL, demonstrating that both arise from a common self-correcting mechanism with task-specific acquisition objectives. We further extend the approach to multi-objective optimization via random scalarization that preserves the single-optimizer conditioning structure. Thorough numerical tests on synthetic and real-world benchmarks across single-objective optimization, multi-objective optimization, and active learning demonstrate that (i) KENDO-BO achieves competitive or superior optimization performance compared to state-of-the-art methods while reducing computational overhead by up to $5\times$ and (ii) KENDO-AL achieves superior predictive calibration over MCMC-based active learning baselines with up to $27\times$ speedup.
Heng Zhang, Haotian Xiang, Qin Lu et al.· 0 citations
Optizing Your Sampling (OYS), which instead treats timestep selection as a black-box optimization problem, optimizing the target metric directly with Bayesian optimization, improves both simple and sophisticated samplers such as Euler and DPM-Solver++.
Travis Zhang, Christian K. Belardi, Justin Lovelace et al.· 0 citations
Self-driving laboratories promise to compress materials-discovery timelines from years to weeks by replacing trial-and-error experimentation with closed-loop, algorithm-guided campaigns. Yet, despite the rapid proliferation of robotic and automation hardware, today's autonomous labs rely almost exclusively on Bayesian optimization (BO) to decide what experiment to run next. BO is a sensible approach to low-dimensional optimization problems with smooth response surfaces, but it struggles in precisely the regimes that matter most for real materials campaigns: tight experimental budgets, dozens of process parameters, mixed-integer choices, hard physical constraints, and noisy expensive measurements. In this talk, I will show how moving from BO to partitioning-based algorithms can substantially improve data efficiency, scale gracefully to dozens of process variables, and handle the constraints and noise that characterize realistic experimental campaigns. I will summarize a recently completed large-scale black-box optimization (BBO) benchmark in which we compared 42 solvers across 502 problems ranging from one to 300 dimensions and from smooth and convex to nonsmooth and nonconvex. The results overturn several community assumptions: BO solves only about 9% of problems within a 2,500-evaluationbudget, while a new branch-and-model (BAM) algorithm reaches an 81% success rate, with GLCCLUSTER, MULTIMIN, MCS, and SNOBFIT also performing strongly. A minimal, irreducible set of eight complementary solvers attains 88% solvability on the full suite. I will then move from in-silico benchmarks to the wet lab, presenting a recent algorithmguided experimental campaign on high-performance perovskite solar cells in which a non- BO solver was used to co-optimize six process variables spanning the perovskite, electrontransport, and hole-transport layers. Time permitting, I will also share early results from applying ensembles of BBO algorithms to digital twins of self-driving labs across additional materials systems. I will close with a forward-looking research vision: accelerating autonomous labs by developing, benchmarking, and experimentally validating data-efficient adaptive algorithms across batteries, semiconductors, catalysts, polymeric membranes, and biomolecules. The benchmarking software will be released as open source, with BAM and most BBO software available free to academic users, so that experimental groups can deploy these tools on their own self-driving platforms. Bio: Nick Sahinidis is the Butler Family Chair and Professor in the H. Milton Stewart School of Industrial and Systems Engineering and the School of Chemical and Biomolecular Engineering at Georgia Tech. His current research activities are at the interface between computer science and operations research, with applications in various engineering and scientific areas, including: global optimization of mixed-integer nonlinear programs: theory, algorithms, and software; informatics problems in chemistry and biology; process and energy systems engineering. Professor Sahinidis teaches mathematical optimization, process systems engineering, and scientific computing. He has developed a bioinformatics M.S. program and has taught courses ranging from thermodynamics and metabolic engineering to approximation algorithms and GPU computing. Sahinidis has served on the editorial boards of many leading journals and in various positions within AIChE (American Institute of Chemical Engineers). He received an NSF CAREER award, the INFORMS Computing Society Prize, the MOS Beale-Orchard-Hays Prize, the Computing in Chemical Engineering Award, the Constantin Carathéodory Prize, and the National Award and Gold Medal from the Hellenic Operational Research Society. Sahinidis is a member of the U.S. National Academy of Engineering and a fellow of AIChE and INFORMS.
N. Sahinidis· Proceedings of the 3rd Found...· 0 citations