Active Learning with Statistical Models

1996/03/01 by D. A. Cohn, Z. Ghahramani, M. I. Jordan · 5 citations
Computer Science · #cs.AI

paper · pdf

published as Journal of Artificial Intelligence Research, Vol 4, (1996), 129-145 · See http://www.jair.org/ for any accompanying files

arxiv created 1996/03/01 · arxiv updated 2014/11/17

Abstract

For many types of machine learning algorithms, one can compute the statistically `optimal' way to select training data. In this paper, we review how optimal data selection techniques have been used with feedforward neural networks. We then show how the same principles may be used to select data for two alternative, statistically-based learning architectures: mixtures of Gaussians and locally weighted regression. While the techniques for neural networks are computationally expensive and approximate, the techniques for mixtures of Gaussians and locally weighted regression are both efficient and accurate. Empirically, we observe that the optimality criterion sharply decreases the number of training examples the learner needs in order to achieve good performance.

Cited by

• Fast and Robust: Task Sampling with Posterior and Diversity Synergies for Adaptive Decision-Makers in Randomized Environments
• Dynamically Allocating Evaluation Effort for Model Ranking
• Machine learning-led semi-automated medium optimization reveals salt as key for flaviolin production in Pseudomonas putida
• Parsimonious Dataset Construction for Laparoscopic Cholecystectomy Structure Segmentation
• Conformalized Generative Bayesian Imaging: An Uncertainty Quantification Framework for Computational Imaging