Toward Optimal Probabilistic Active Learning Using a Bayesian Approach
This work addresses the challenge of costly data labeling for machine learning practitioners, offering an incremental improvement over existing active learning methods.
The authors tackled the problem of reducing labeling costs in machine learning by proposing a decision-theoretic active learning strategy that directly optimizes misclassification error gain and uses a Bayesian approach to handle uncertainties, achieving superior performance validated through extensive experiments on various datasets and kernels.
Gathering labeled data to train well-performing machine learning models is one of the critical challenges in many applications. Active learning aims at reducing the labeling costs by an efficient and effective allocation of costly labeling resources. In this article, we propose a decision-theoretic selection strategy that (1) directly optimizes the gain in misclassification error, and (2) uses a Bayesian approach by introducing a conjugate prior distribution to determine the class posterior to deal with uncertainties. By reformulating existing selection strategies within our proposed model, we can explain which aspects are not covered in current state-of-the-art and why this leads to the superior performance of our approach. Extensive experiments on a large variety of datasets and different kernels validate our claims.