Exploration vs. exploitation in active learning : A Bayesian approach

Bondu, Alexis; Lemaire, Vincent; Boullé, Marc

doi:10.1109/ijcnn.2010.5596815

Cited by 29 publications

(23 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore an explorationexploitation method with a large proportion of random sampling during the early queries is adopted. Some authors also consider combining different criteria [9], [10] or selecting the strategies adaptively for a better performance. [8], [11], [12] perform adaptive strategy selection by connecting the selection problem to multi-arm bandit methods.…”

Section: B Active Learningmentioning

confidence: 99%

Active Learning for High-Dimensional Binary Features

Vahdat

Belbahri

Nia

2019

2019 15th International Conference on Network and Service Management (CNSM)

View full text Add to dashboard Cite

Erbium-doped fiber amplifier (EDFA) is an optical amplifier/repeater device used to boost the intensity of optical signals being carried through a fiber optic communication system. A highly accurate EDFA model is important because of its crucial role in optical network management and optimization. The input channels of an EDFA device are treated as either on or off, hence the input features are binary. Labeled training data is very expensive to collect for EDFA devices, therefore we devise an active learning strategy suitable for binary variables to overcome this issue. We propose to take advantage of sparse linear models to simplify the predictive model. This approach simultaneously improves prediction and accelerates active learning query generation. We show the performance of our proposed active learning strategies on simulated data and real EDFA data.

show abstract

Section: B Active Learningmentioning

confidence: 99%

Active Learning for High-Dimensional Binary Features

Vahdat

Belbahri

Nia

2019

2019 15th International Conference on Network and Service Management (CNSM)

View full text Add to dashboard Cite

show abstract

“…[3,6,23]), which is central in motivating our approach. As an example for algorithms that conduct merely refinement, we consider 'SIMPLE' [28].…”

Section: Related Workmentioning

confidence: 99%

“…In a slightly different view, it is equivalent to the iterative application of the kernel (1) on the characteristic vector χ, while restarting χ l to Y l after each iteration. Its probabilistic justification is in the random walk framework (6), which leads to classification. The summary of the label diffusion algorithm DiffuseLabels is given in Fig.…”

Section: Classification Via Diffusion Kernelsmentioning

confidence: 99%

See 1 more Smart Citation

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

“…So, the initial sets of feedback play a significant role to the model's configuration hence, the model may risks suffering from the positive reinforcement phenomenon. In other words, high influence of the current model to select patterns for obtaining the user's feedback in the next iteration is similar to exploitation in active learning [7]. It sometimes makes the predictive model represents some areas of the pattern space, making it sub-optimal over the entire pattern space.…”

Section: Introductionmentioning

confidence: 99%

PRIIME: A generic framework for interactive personalized interesting pattern discovery

Bhuiyan

Hasan

2016

2016 IEEE International Conference on Big Data (Big Data)

View full text Add to dashboard Cite

The traditional frequent pattern mining algorithms generate an exponentially large number of patterns of which a substantial proportion are not much significant for many data analysis endeavors. Discovery of a small number of personalized interesting patterns from the large output set according to a particular user's interest is an important as well as challenging task. Existing works on pattern summarization do not solve this problem from the personalization viewpoint. In this work, we propose an interactive pattern discovery framework named PRIIME which identifies a set of interesting patterns for a specific user without requiring any prior input on the interestingness measure of patterns from the user. The proposed framework is generic to support discovery of the interesting set, sequence and graph type patterns. We develop a softmax classification based iterative learning algorithm that uses a limited number of interactive feedback from the user to learn her interestingness profile, and use this profile for pattern recommendation. To handle sequence and graph type patterns PRIIME adopts a neural net (NN) based unsupervised feature construction approach. We also develop a strategy that combines exploration and exploitation to select patterns for feedback. We show experimental results on several real-life datasets to validate the performance of the proposed method. We also compare with the existing methods of interactive pattern discovery to show that our method is substantially superior in performance. To portray the applicability of the framework, we present a case study from the real-estate domain.

show abstract

Exploration vs. exploitation in active learning : A Bayesian approach

Cited by 29 publications

References 7 publications

Active Learning for High-Dimensional Binary Features

Active Learning for High-Dimensional Binary Features

Active-transductive learning with label-adapted kernels

PRIIME: A generic framework for interactive personalized interesting pattern discovery

Contact Info

Product

Resources

About