Transductive Active Learning: Theory and Applications
This addresses the challenge of efficient data sampling for machine learning tasks where target predictions extend beyond accessible data regions, with applications in neural network fine-tuning and Bayesian optimization.
The paper tackles the problem of active learning in real-world settings where sampling is restricted to an accessible region but prediction targets may lie outside, by analyzing decision rules that adaptively sample to minimize uncertainty. It shows that these rules converge uniformly to minimal uncertainty and demonstrates strong sample efficiency, achieving state-of-the-art performance in applications like active fine-tuning of large neural networks and safe Bayesian optimization.
We study a generalization of classical active learning to real-world settings with concrete prediction targets where sampling is restricted to an accessible region of the domain, while prediction targets may lie outside this region. We analyze a family of decision rules that sample adaptively to minimize uncertainty about prediction targets. We are the first to show, under general regularity assumptions, that such decision rules converge uniformly to the smallest possible uncertainty obtainable from the accessible data. We demonstrate their strong sample efficiency in two key applications: active fine-tuning of large neural networks and safe Bayesian optimization, where they achieve state-of-the-art performance.