Repairing self-confident active–transductive learners using systematic exploration

Begleiter, Ron; El‐Yaniv, Ran; Pechyony, Dmitry

doi:10.1016/j.patrec.2008.01.025

Cited by 1 publication

(10 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, and unlike other active classifiers (e.g. [3,23,35]), our algorithm does not require a separate run of several classifiers for each query, and does not employ computationally demanding clustering algorithms, such as [4,22]. It exhibits orders of magnitude speedup in empirical experiments, confirming its asymptotic complexity, while achieving superior accuracy.…”

Section: Introductionmentioning

confidence: 78%

“…However, the implementation is still demanding in terms of running time. As a third example we consider the algorithm +EXPLORE of [4]. The authors use spectral clustering [21] to expose uncovered clusters that the learner missed while querying samples.…”

Section: Related Workmentioning

confidence: 99%

“…For comparison, we discuss the running time of a few relevant active learners: Both [4] and [22] employ full clustering schemes for each query. '+ EXPLORE' [4] involves a spectral algorithm whose computational effort in general is cubic in N .…”

Section: Running Time Of the Algorithmmentioning

confidence: 99%

“…'+ EXPLORE' [4] involves a spectral algorithm whose computational effort in general is cubic in N . As the authors report they had to half the size of standard benchmark UCI data sets in order to reduce a running time of over a month with 20 CPUs.…”

Section: Running Time Of the Algorithmmentioning

confidence: 99%

“…A second approach is a combined one, which employs both refinement of the version space and general exploration of the feature space (e.g. [3,4,6,23]). These methods typically use graph-based techniques, where the exploration step samples pockets of data that the learner misclassifies and refines their class boundaries.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Active-transductive learning with label-adapted kernels

Kushnir¹

2014

Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

View full text Add to dashboard Cite

This paper presents an efficient active-transductive approach for classification. A common approach of active learning algorithms is to focus on querying points near the class boundary in order to refine it. However, for certain data distributions, this approach has been shown to lead to uninformative samples. More recent approaches consider combining data exploration with traditional refinement techniques. These techniques typically require tuning sampling of unexplored regions with refinement of detected class boundaries. They also involve significant computational costs for the exploration of informative query candidates. We present a novel iterative active learning algorithm designed to overcome these shortcomings by using a linear running-time activetransductive learning approach that naturally switches from exploration to refinement. The passive classifier employed in our algorithm builds a random-walk on the data graph based on a modified graph geometry that combines the data distribution with current label hypothesis; while the query component uses the uncertainty of the evolving hypothesis. Our supporting theory draws the link between the spectral properties of our iteration matrix and a solution to the minimal-cut problem for a fused hypothesis-data graph. Experiments demonstrate computational complexity that is orders of magnitude lower than state-of-the-art, and competitive results on benchmark data and real churn prediction data.

show abstract

Section: Introductionmentioning

confidence: 78%

Section: Related Workmentioning

confidence: 99%