Toward Optimal Active Learning through Monte Carlo Estimation of Error Reduction

By Nicholas A. Roy, Andrew McCallum. 2001

397 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Reliable Uncertainty Estimates in Deep Neural Networks using Noise Contrastive Priors (2018, methodology): “proaches use classifier confidence (Lewis and Gale, 1994), predicted expected error (Roy and McCallum, 2001), or model disagreement (Houlsby et al.”
Ask the Right Questions: Active Question Reformulation with Reinforcement Learning (2017, methodology): “1Bayes optimal active learning also tries to optimize end-to-end performance, however, these techniques are often computationally prohibitive in large-scale settings [Roy and McCallum, 2001]”
Learning how to Active Learn: A Deep Reinforcement Learning Approach (2017, background): “, 1992), which selects the data about which the “committee” disagree most; and 3) expected error reduction (Roy and McCallum, 2001), which selects the data that can contribute the largest model loss reduction for the current classifier once labelled.”
Learning Loss for Active Learning (2019, methodology): “Expected model change [44, 48, 12] selects data points that would cause the greatest change to the current model parameters or outputs if we knew their labels.”
Active Learning for Convolutional Neural Networks: A Core-Set Approach (2017, methodology): “Bayesian active learning methods typically use a non-parametric model like Gaussian process to estimate the expected improvement by each query (Kapoor et al., 2007) or the expected error after a set of queries (Roy & McCallum, 2001).”
Explanatory Interactive Machine Learning (2019, methodology): “To this end, the query instances are chosen to be as informative as possible, typically by maximizing some informativeness criterion, such as the expected model improvement [28] or practical approximations thereof.”
Trajectory Optimization using Reinforcement Learning for Map Exploration (2008, background): “Using Naive Bayesian independence assumptions can often allow computationally efficient and accurate inference of the mean of a distribution but the higher order moments are generally biased and over-confident, leading to problems in active learning and exploration (Roy and McCallum 2001).”
Variational Adversarial Active Learning (2019, methodology): “In the realm of Bayesian frameworks, probabilistic models such as Gaussian processes [25, 41] or Bayesian neural networks [9] are used to estimate uncertainty.”
Semi-Automated Data Labeling (2020, background): “Popular selection heuristics include uncertainty, expected error reduction Roy and McCallum (2001) and query by committee Freund et al. (1997).”
A Survey of Deep Active Learning (2020, background): “For example, in a given set of unlabeled datasets, the main query strategies include the uncertaintybased approach [19, 106, 131, 173, 196, 222], diversity-based approach [23, 72, 84, 153], and expected model change [69, 177, 195].”
Textual Data Augmentation for Efficient Active Learning on Tiny Datasets (2020, background): “Finally, expected-model change selects examples that would cause the greatest change to a model’s output if their labels were known (Freytag et al., 2014; Roy and McCallum, 2001; Settles et al., 2008).”
Active Learning for Visual Question Answering: An Empirical Study (2017, background): “Previous works explore the goal-driven learning objective for shallow classiﬁers such as Naive Bayes [29], Support Vector Machines [12] and Gaussian Process [35].”