Variational Adversarial Active Learning

By Samarth Sinha, Sayna Ebrahimi, Trevor Darrell. 2019

680 citations recorded by the source corpus.

Active learning aims to develop label-efficient algorithms by sampling the most representative queries to be labeled by an oracle. We describe a pool-based semi-supervised active learning algorithm that implicitly learns this sampling mechanism in an adversarial manner. Our method learns a latent space using a variational autoencoder (VAE) and an adversarial network trained to discriminate between unlabeled and labeled data. The mini-max game between the VAE and the adversarial network is played such that while the VAE tries to trick the adversarial network into predicting that all data points are from the labeled pool, the adversarial network learns how to discriminate between dissimilarities in the latent space. We extensively evaluate our method on various image classification and semantic segmentation benchmark datasets and establish a new state of the art on CIFAR10/100, Caltech-256, ImageNet, Cityscapes, and BDD100K. Our results demonstrate that our adversarial approach learns an effective low dimensional latent space in large-scale settings and provides for a computationally efficient sampling method. Our code is available at \url{https://github.com/sinhasam/vaal}.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Adversarial Continual Learning (2020, methodology): “Adversarial learning of the latent space has been extensively researched in domain adaptation [14], active learning [34], and representation learning [17,22].”
A Wholistic View of Continual Learning with Deep Neural Networks: Forgotten Lessons and the Bridge to Active and Open World Learning (2020, background): “Variational adversarial active learning (VAAL) (Sinha et al., 2019) approximates the data distribution by learning the latent space in a variational autoencoder (Kingma and Ba, 2015) and simultaneously trains a latent based adversarial network to discriminate between unlabelled and labelled data.”
GroupViT: Semantic Segmentation Emerges from Text Supervision (2022, methodology): “For example, few-shot learning [21, 43, 50, 55, 70, 77, 84] and active learning [8, 62, 66, 67, 83] are proposed to perform segmentation with as few pixel-wise labels as possible.”
Learning to Predict Gradients for Semi-Supervised Continual Learning (2022, methodology): “This gradient-based learning process is key to updating the models to make more precise predictions [5]–[10], [12]–[15].”
Towards Robust and Reproducible Active Learning using Neural Networks (2020, methodology): “Abbreviations: Active Learning (AL), Random Sampling Baseline (RSB), Query-by-Committee (QBC), Variational Adversarial Active Learning (VAAL), Uncertainty-based sampling (UC), Deep Bayesian Active Learning (DBAL), Bayesian Active Learning by Disagreement (BALD), Random Augmentation (RA), Stochastic ”
A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges (2020, methodology): “24: A schematic view of the VAAL model which is reproduced based on [488].”
Textual Data Augmentation for Efficient Active Learning on Tiny Datasets (2020, background): “To address this issue, some works incorporate measures to exploit the diversity information of the examples in the selection process (Sener and Savarese, 2017; Wang et al., 2017; Sinha et al., 2019).”
A Survey of Deep Active Learning (2020, methodology): “(a) Structure comparison chart of VAAL [150] and TA-VAAL [81].”
BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning (2019, background): “Wang et al. [35], Sener and Savarese [29], Samarth Sinha [28] explore combining it with AL.”
Consistency Regularization for Variational Auto-Encoders (2021, background): “(2015), active learning (Sinha et al., 2019), continual learning (Achille et al.”