Importance weighted active learning

By Alina Beygelzimer, Sanjoy Dasgupta, J. Langford. 2008

371 citations recorded by the source corpus.

We present a practical and statistically consistent scheme for actively learning binary classifiers under general loss functions. Our algorithm uses importance weighting to correct sampling bias, and by controlling the variance, we are able to give rigorous label complexity bounds for the learning process.

Subfields: Artificial Intelligence

How researchers see this paper

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

In Artificial Intelligence

Citations in the corpus

The Unreasonable Effectiveness of Noisy Data for Fine-Grained Recognition (2015, neutral): “In contrast to our work, the oracle is often simulated with either a simple or no noise model [4, 32, 10, 55].”
Deep Active Learning for Named Entity Recognition (2017, background): “Active learning While learning-theoretic properties of active learning algorithms are wellstudied (Dasgupta et al., 2005; Balcan et al., 2009; Awasthi et al., 2014; Yan and Zhang, 2017; Beygelzimer et al., 2009), classic algorithms and guarantees cannot be generalized to DNNs, which are currently ar”
Active Learning Literature Survey (2009, background): “However, some recent theoretical work has begun to address these issues, coupled with promising empirical results (Dasgupta and Hsu, 2008; Beygelzimer et al., 2009).”
Online Importance Weight Aware Updates (2010, background): “∗Research done while author was at Yahoo! Research to the test distribution the example is, and active learning algorithms [1, 2] where an adaptive rejection sampling scheme is applied to each example and each retained example gets an importance equal to the inverse probability of being retained.”
A theory of transfer learning with applications to active learning (2012, background): “…individuals, a learning sys-tem can identify common patterns of speech, such as accents or dialects, each of which requires a slightly different speech recognizer; then, given a new person to train a recognizer for, it can quickly determine the particular dialect from only a few well-chosen…”
Generative Adversarial Active Learning (2017, neutral): “Typical meth-ods include (Beygelzimer et al., 2008; Cohn et al., 1994; Dasgupta et al., 2007).”
PAC-Bayes-Bernstein Inequality for Martingales and its Application to Multiarmed Bandits (2011, methodology): “The existing theoretical analysis of the exploration-exploitation trade-off in problems that go beyond multiarmed bandits is mainly focused on the worst-case scenarios (Strehl et al., 2009; Jaksch et al., 2010; Beygelzimer et al., 2011, 2009).”
Rates of convergence in active learning (2011, background): “See [8, 10, 14, 16, 19, 33] for further discussions of various uses of the disagreement coefficient and related notions and extensions in active learning.”
Easy Questions First? A Case Study on Curriculum Learning for Question Answering (2016, methodology): “2) Importance-Weighting (IW): Importance weighting is a common technique in active learning literature (Tong and Koller, 2002; Beygelzimer et al., 2009; Beygelzimer et al., 2010), which”
Agnostic Selective Classification (2011, methodology): “Our results can be extended to losses other than 0/1 by similar techniques to those used in [10].”
Crowdsourced Data Management: Industry and Academic Perspectives (2015, background): “Papers focusing on the theory of active learning [56, 64, 111, 31, 92, 38, 37, 28, 63, 198, 136] show that the adaptive selection strategies proposed (i.”
A survey on instance selection for active learning (2012, background): “Two most successful types are: (1) Semi-supervised learning [87,88], which directly utilizes unlabeled instances by taking into account the geometry of data distributions, such as clusters and manifolds, to propagate label information to neighboring data; and (2) Active learning [89,90], which selec”
Alignment for Advanced Machine Learning Systems (2020, background): “Seung, Opper, and Sompolinsky (1992) and Beygelzimer, Dasgupta, and Langford (2009) have both studied what statistical guarantees can be achieved in this setting.”
Semi-Supervised Active Learning for Sound Classification in Hybrid Learning Environments (2016, background): “While the current fixed threshold strategy may not be suitable in other classification tasks, one can refer to [47], [48] and the references therein for more sophisticated thresholding and selection criteria that delicately balance the trade-off between asking for human labeling versus receiving…”
Active Learning as Non-Convex Optimization (2009, methodology): “Recently, Beygelzimer et al. (2008) presented the first practical, consistent active learning algorithm that gives label complexity improvements for convex losses although these improvements are not as good as in the zero-one loss case and require the loss to satisfy certain properties.”
What is the Effect of Importance Weighting in Deep Learning? (2018, background): “One related use of IW-ERM is to correct for sampling bias in active learning (Beygelzimer et al., 2009; Settles, 2010).”
A compression technique for analyzing disagreement-based active learning (2014, neutral): “…literature (see e.g., Balcan, Beygelzimer, and Langford, 2009; Hanneke, 2007a; Dasgupta, Hsu, and Monteleoni, 2007; Balcan, Broder, and Zhang, 2007; Beygelzimer, Dasgupta, and Langford, 2009; Friedman, 2009; Balcan, Hanneke, and Vaughan, 2010; Hanneke, 2011; Koltchinskii, 2010; Beygelzimer, Hsu,…”