Active and Adaptive Sequential learning
This addresses the problem of efficiently learning from sequentially evolving data streams for machine learning practitioners, representing an incremental improvement to existing active learning methods.
The authors developed an active learning framework for sequentially changing machine learning problems, where their algorithm actively queries informative samples and adapts to changes using previous information. Their analysis shows it achieves near-optimal excess risk performance for maximum likelihood estimation and provides an adaptive sample size rule that bounds excess risk over time.
A framework is introduced for actively and adaptively solving a sequence of machine learning problems, which are changing in bounded manner from one time step to the next. An algorithm is developed that actively queries the labels of the most informative samples from an unlabeled data pool, and that adapts to the change by utilizing the information acquired in the previous steps. Our analysis shows that the proposed active learning algorithm based on stochastic gradient descent achieves a near-optimal excess risk performance for maximum likelihood estimation. Furthermore, an estimator of the change in the learning problems using the active learning samples is constructed, which provides an adaptive sample size selection rule that guarantees the excess risk is bounded for sufficiently large number of time steps. Experiments with synthetic and real data are presented to validate our algorithm and theoretical results.