Improving Generalization with Active Learning

By David A. Cohn, Les E. Atlas, Richard E. Ladner. 1994

Subfields: Meta & Continual Learning, Statistical Learning Theory

How researchers see this paper

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

In Meta & Continual Learning

In Statistical Learning Theory

Citations in the corpus

Rewriting a Deep Generative Model (2020, background): “Interactive machine learning systems aim to improve training through human interaction in labeling [15,22,68], or by allowing a user to to aid in the model optimization process via interactive feature selection [19,27,64,50] or model”
Scalable Private Learning with PATE (2018, background): “In Machine Learning Theory, active learning (Cohn et al., 1994) has been shown to allow learning from fewer labeled examples than the passive case (see e.g. Hanneke (2014)).”
Deep Active Learning with a Neural Architecture Search (2018, background): “The literature on active learning in the context of classical models such as SVMs is extensive [3, 6, 27, 2, 1, 15], and clearly beyond the scope of this paper.”
Replay in Deep Learning: Current Approaches and Missing Biological Elements (2021, neutral): “The sample selection problem is also closely related to active learning strategies (Cohn, Atlas, & Ladner, 1994; Lin & Parikh, 2017; Wang, Zhang, Li, Zhang, & Lin, 2016; Settles, 2009; Yoo & So Kweon, 2019; Wei, Iyer, & Bilmes, 2015), with the most common selection methods using uncertainty…”
Mastering the Dungeon: Grounded Language Learning by Mechanical Turker Descent (2017, neutral): “Choosing the best examples for the model to see next is also related to active learning (Cohn et al., 1994) except in our case this is chosen on the teacher’s, rather than the learner’s side.”
End-to-end LSTM-based dialog control optimized with supervised and reinforcement learning (2016, background): “We next examined whether the model would be suitable for active learning (Cohn et al., 1994).”
Domain Adaptive Dialog Generation via Meta Learning (2019, background): “Transfer learning (Caruana, 1997a; Bengio, 2012; Cohn et al., 1994; Mo et al., 2018), few-shot learning (Salakhutdinov et al.”
A General Active-Learning Framework for On-Road Vehicle Recognition and Tracking (2010, methodology): “The query function is central to the active-learning process [20] and generally serves to select difficult training examples, which are informative in updating a decision boundary [6].”
Active Learning with Feedback on Features and Instances (2006, methodology): “Techniques such as active learning (Cohn et al., 1994), semi-supervised learning (Zhu, 2005), and transduction (Joachims, 1999) have been pursued with considerable success in reducing labeling requirements.”
Knows what it knows: a framework for self-aware learning (2008, background): “KWIK could also be a useful framework for studying active learning (Cohn et al., 1994) and anomaly detection (Lane & Brodley, 2003), both of which are machine-learning problems that require some degree of reasoning about whether a recently presented input is predictable from previous…”
A Survey on Active Learning and Human-in-the-Loop Deep Learning for Medical Image Analysis (2019, background): “Stream-based Selective Sampling assumes a continuous stream of incoming un-annotated data-points x U (Atlas et al. (1990); Cohn et al. (1994)).”
Decoupling "when to update" from "how to update" (2017, background): “In other works, an ensemble of neural networks is trained in an active learning setup to improve the generalization of neural networks [11, 2, 22].”
SHIV: Reducing supervisor burden in DAgger using support vectors for efficient learning from demonstrations in high dimensional state spaces (2016, background): “Active Learning from Demonstration We formulate reducing supervisor burden in online Lfd as a stream-based active learning problem [5], [9].”
Multitask Learning for Spoken Language Understanding (2006, background): “This is done by inspecting the unlabeled examples and selecting the most informative ones, with respect to a given cost function , for a human to label [8].”
The Relationship Between Agnostic Selective Classification Active Learning and the Disagreement Coefficient (2017, methodology): “The first formal relationship between PCS classification and AL was proposed in [20, 21], where the aforementioned CSS algorithm was shown to be equivalent to the well-known CAL AL algorithm of [22], in the sense that a fast coverage rate for CSS was proven to be equivalent to exponential label comp”