This paper explores the application of optimal data selection techniques to active learning, focusing on feedforward neural networks, mixtures of Gaussians, and locally weighted regression. The authors review how optimal data selection can be applied to neural networks, which are computationally expensive and approximate, and demonstrate that for mixtures of Gaussians and locally weighted regression, the techniques are both efficient and accurate. Empirical results show that the optimality criterion significantly reduces the number of training examples needed to achieve good performance. The paper also discusses the computational costs and potential savings in data acquisition, particularly in industrial settings where data collection is costly. The authors conclude by highlighting the potential of these methods for active learning in various applications, including function approximation and classification problems.This paper explores the application of optimal data selection techniques to active learning, focusing on feedforward neural networks, mixtures of Gaussians, and locally weighted regression. The authors review how optimal data selection can be applied to neural networks, which are computationally expensive and approximate, and demonstrate that for mixtures of Gaussians and locally weighted regression, the techniques are both efficient and accurate. Empirical results show that the optimality criterion significantly reduces the number of training examples needed to achieve good performance. The paper also discusses the computational costs and potential savings in data acquisition, particularly in industrial settings where data collection is costly. The authors conclude by highlighting the potential of these methods for active learning in various applications, including function approximation and classification problems.