Abstract:In many applications, data is easy to acquire but expensive and time-consuming to label prominent examples include medical imaging and NLP. This disparity has only grown in recent years as our ability to collect data improves. Under these constraints, it makes sense to select only the most informative instances from the unlabeled pool and request an oracle (e.g., a human expert) to provide labels for those samples. The goal of active learning is to infer the informativeness of unlabeled samples so as to minimize the number of requests to the oracle. Here, we formulate active learning as an open-set recognition problem. In this paradigm, only some of the inputs belong to known classes; the classifier must identify the rest as unknown. More specifically, we leverage variational neural networks (VNNs), which produce high-confidence (i.e., low-entropy) predictions only for inputs that closely resemble the training data. We use the inverse of this confidence measure to select the samples that the oracle should label. Intuitively, unlabeled samples that the VNN is uncertain about are more informative for future training. We carried out an extensive evaluation of our novel, probabilistic formulation of active learning, achieving state-of-the-art results on MNIST, CIFAR-10, and CIFAR-100. Additionally, unlike current active learning methods, our algorithm can learn tasks without the need for task labels. As our experiments show, when the unlabeled pool consists of a mixture of samples from multiple datasets, our approach can automatically distinguish between samples from seen vs. unseen tasks.

Active Learning with Extremely Sparse Labeled Examples.

High Reliable Multi-View Semi-Supervised Learning with Extremely Sparse Labeled Data

Exemplar Guided Active Learning

Uncertainty-Based Active Learning Via Sparse Modeling for Image Classification

Active Learning from Weak and Strong Labelers

Column Subset Selection for Active Learning in Image Classification

Learning Distinctive Margin Toward Active Domain Adaptation

Active Learning: Problem Settings and Recent Developments

Using methods from dimensionality reduction for active learning with low query budget

Unlabeled data selection for active learning in image classification

Cost-Effective Active Learning from Diverse Labelers.

Cost Effective Multi-label Active Learning Via Querying Subexamples

Active Learning with Label Quality Control

Training Ensembles with Inliers and Outliers for Semi-supervised Active Learning

Active Learning with Label Correlation Exploration for Multi-Label Image Classification

Active subspace learning

Semi-supervised learning with very few labeled training examples

Active WeaSuL: Improving Weak Supervision with Active Learning

Deep Active Learning via Open Set Recognition

Adversarial Active Learning for Deep Networks: a Margin Based Approach

An efficient active Bayes classifier using affinity propagation on unlabeled data