Dataefficient

Dataefficient, or data-efficient, is a term used in machine learning and data science to describe methods, models, or systems that achieve competitive or superior performance while using relatively small amounts of labeled data, or that extract maximum information from available data. The emphasis is on sample efficiency—the amount of data required to reach a given level of performance—rather than on raw model size or computational throughput alone.

In practice, data efficiency contrasts with data-hungry approaches that require vast labeled datasets. Data efficiency can apply to labeled data, unlabeled data, or both, and often involves leveraging high-quality priors, structured representations, or external information to reduce data requirements.

Common techniques associated with data-efficient learning include semi-supervised and self-supervised learning to utilize unlabeled data, few-shot and meta-learning to adapt quickly from limited examples, transfer learning to reuse knowledge from related tasks, active learning to selectively label informative samples, and data augmentation or synthetic data generation to expand effective training sets. Architectural choices, regularization, and probabilistic modeling can also improve data efficiency by making learning more robust to limited data and by better exploiting existing information.

Data efficiency is relevant across domains such as computer vision, natural language processing, robotics, and reinforcement learning, where sample acquisition can be costly or time-consuming. It is often evaluated through learning curves and metrics that reflect performance per sample, with ongoing research aimed at improving data efficiency without sacrificing generalization or fairness.