Advances in science and engineering increasingly depend on machine learning models to accelerate discovery, design, and decision-making. Yet, many applications operate in low-data regimes, where sample acquisition is costly, time-intensive, or experimentally constrained [1]. Some representative cases include generating high-fidelity datasets for turbulence closure model calibration, geometrical design optimization, and the development of digital twins of biological systems with inherently inaccessible empirical data. In such settings, the ability to extract maximal information per sample becomes critical. This mini symposium will address the central challenge of how to learn effectively when data is scarce. Topics include adaptive experiment design, active learning, uncertainty quantification, and optimization, as well as surrogate modeling architectures that integrate known (or partially known) physics, use multi-level and multi-fidelity data or employ reduced-order approaches for efficient preprocessing [2]. A particular focus will be on models that adapt their sampling strategies as they learn, iteratively refining surrogates to improve accuracy and robustness even under sparse or noisy measurement conditions. By combining intelligent sampling with cost-aware model design, these approaches enable fast-to-train, generalizable surrogates suitable for deployment in environments where reducing model uncertainty and computational efficiency are essential. REFERENCES: [1] U. Fasel, J. N. Kutz, B. W. Brunton, and S. L. Brunton. Ensemble-sindy: Robust sparse model discovery in the low-data, high-noise limit, with active learning and control. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 478(2260):20210904, 2022. [2] Fu, Y., Zhu, X. & Li, B. A survey on instance selection for active learning. Knowl Inf Syst 35, 249–283, 2013.