Large-Scale Visual Active Learning with Deep Probabilistic Ensembles

Abstract

Annotating the right data for training deep neural networks is an important challenge. Active learning using uncertainty estimates from Bayesian Neural Networks (BNNs) could provide an effective solution to this. Despite being theoretically principled, BNNs require approximations to be applied to large-scale problems, where both performance and uncertainty estimation are crucial. In this paper, we introduce Deep Probabilistic Ensembles (DPEs), a scalable technique that uses a regularized ensemble to approximate a deep BNN. We conduct a series of large-scale visual active learning experiments to evaluate DPEs on classification with the CIFAR-10, CIFAR-100 and ImageNet datasets, and semantic segmentation with the BDD100k dataset. Our models require significantly less training data to achieve competitive performances, and steadily improve upon strong active learning baselines as the annotation budget is increased.

arXiv commentarXiv admin note: text overlap with arXiv:1811.02640

References in corpus (15)

• Improving neural networks by preventing co-adaptation of feature detectors
• How transferable are features in deep neural networks?
• What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
• On Calibration of Modern Neural Networks
• Weight Uncertainty in Neural Networks
• Bayesian Convolutional Neural Networks with Bernoulli Approximate Variational Inference
• Deep Bayesian Active Learning with Image Data
• Identity Mappings in Deep Residual Networks
• Bayesian Uncertainty Estimation for Batch Normalized Deep Networks
• Model-Based Active Exploration
• Active One-shot Learning
• Meta-Learning Transferable Active Learning Policies by Deep Reinforcement Learning
• Learning how to Active Learn: A Deep Reinforcement Learning Approach
• Sufficient Conditions for Idealised Models to Have No Adversarial Examples: a Theoretical and Empirical Study with Bayesian Neural Networks
• Adaptive Semantic Segmentation with a Strategic Curriculum of Proxy Labels

Cited by in corpus (5)

• A Review and Comparative Study on Probabilistic Object Detection in Autonomous Driving
• Diminishing Uncertainty within the Training Pool: Active Learning for Medical Image Segmentation
• Interactive Weak Supervision: Learning Useful Heuristics for Data Labeling
• Deep Active Learning for Efficient Training of a LiDAR 3D Object Detector
• Looking at the posterior: accuracy and uncertainty of neural-network predictions