Analysis of a greedy active learning strategy

By Sanjoy Dasgupta. 2004

375 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

Active learning of inverse models with intrinsically motivated goal exploration in robots (2013, neutral): “n as a goal reached with the highest competence level have important in uences on the learning process. If we look at traditional active learning algorithms which cannot deal with open-ended learning [30, 43, 32], as well as RIAC-like algorithms dierent from SAGG-RIAC [77, 65, 94, 69, 96, 8], we can”
A Wholistic View of Continual Learning with Deep Neural Networks: Forgotten Lessons and the Bridge to Active and Open World Learning (2020, background): “complexity and necessary amount of queries to be analyzed with respect to these binary classification problems with linear decision boundary in the context of greedy active learning strategies [116].”
Supervision via competition: Robot adversaries for learning tasks (2016, background): “In simplified, linear classification problems, it has been shown that exponential decrease is achievable in expectation with no efficiency gains in the worst case [13].”
Bayesian Active Learning for Classification and Preference Learning (2011, background): “It has been shown that the myopic policy can perform near-optimally [Golovin and Krause, 2010, Dasgupta, 2005].”
Margin Based Active Learning (2007, background): “We consider here a commonly studied setting in the active learning literature [7–9].”
Model learning for robot control: a survey (2011, background): “The goal is to obtain a sparse and informative labeled training set by intelligent adaptive querying (Cohn et al. 1996; Dasgupta 2004).”
Adversarial Active Learning for Deep Networks: a Margin Based Approach (2018, background): “Several heuristics coexist as it is impossible to obtain a universal active learning strategy effective for any given task (Dasgupta, 2005).”
Query by Committee Made Real (2005, methodology): “Two other notable algorithm s are the greedy active learning algorithm [11] and the perceptron based active learning algorithm [12].”
Upper and Lower Error Bounds for Active Learning (2006, background): “In certain noiseless scenarios it has been shown that the number of labeled examples needed is to achieve a desired classification error rate is much smaller than what would be need using passive learning (in fact for those scenarios if passive learning requires n labeled examples then active learni”
Generative Adversarial Active Learning (2017, background): “ is able to achieve accuracies better than that of the fully supervised scheme. With 350 training samples, its accuracy improves over supervised learning and even SVMactive, an aggressiveactivelearner[11, 45]. Obviously,theaccuracyofbothSVMactive andrandomsampling will eventually converge to the ful”
Analysis of Perceptron-Based Active Learning (2009, background): “For instance, in our setting where data is distributed uniformly over the unit sphere, recent work has shown that if the target linear separator is allowed to be non-homogeneous, then the number of labels required to reach error is Ω(1/ ), no matter what active learning scheme is used [5].”
A theory of transfer learning with applications to active learning (2012, unknown): “, Freund et al. (1997), Dasgupta (2004, 2005), Dasgupta et al.”
Submodular Function Maximization (2014, background): “One example is active learning in the noiseless case ( c.f. , Kosaraju et al. 1999; Dasgupta 2004; Golovin and Krause 2011b), in which we must adaptively select data points to be labelled for us (at some cost) until we can infer the labeling of all data points, while attempting to minimize the cost.”
Guest Editorial Active Learning and Intrinsically Motivated Exploration in Robots: Advances and Challenges (2010, background): “It has been proved that, under mild conditions, an active learner outperforms a passive learner [6], [17], provided that we have some information on the structure of the problem, even when we cannot estimate in advance that improvement [18].”
Active Learning for Convolutional Neural Networks: A Core-Set Approach (2017, background): “On the theoretical side, it is shown that greedy active learning is not possible in algorithm and data agnostic case (Dasgupta, 2005).”