Adversarial Active Learning for Deep Networks: a Margin Based Approach

Abstract

We propose a new active learning strategy designed for deep neural networks. The goal is to minimize the number of data annotation queried from an oracle during training. Previous active learning strategies scalable for deep networks were mostly based on uncertain sample selection. In this work, we focus on examples lying close to the decision boundary. Based on theoretical works on margin theory for active learning, we know that such examples may help to considerably decrease the number of annotations. While measuring the exact distance to the decision boundaries is intractable, we propose to rely on adversarial examples. We do not consider anymore them as a threat instead we exploit the information they provide on the distribution of the input space in order to approximate the distance to decision boundaries. We demonstrate empirically that adversarial active queries yield faster convergence of CNNs trained on MNIST, the Shoe-Bag and the Quick-Draw datasets.

References in corpus (4)

• Deep Bayesian Active Learning with Image Data
• Defensive Distillation is Not Robust to Adversarial Examples
• QBDC: Query by dropout committee for training deep supervised architecture
• Active Decision Boundary Annotation with Deep Generative Models

Cited by in corpus (44)

• Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds
• Discriminative Active Learning
• A Survey of Active Learning for Text Classification using Deep Neural Networks
• Are All Training Examples Created Equal? An Empirical Study
• A critical look at the current train/test split in machine learning
• A framework for the extraction of Deep Neural Networks by leveraging public data
• Deep Active Learning for Computer Vision: Past and Future
• Deep Ensemble Bayesian Active Learning : Addressing the Mode Collapse issue in Monte Carlo dropout via Ensembles
• Label-Efficient Point Cloud Semantic Segmentation: An Active Learning Approach
• Gone Fishing: Neural Active Learning with Fisher Embeddings
• Deep Active Learning: Unified and Principled Method for Query and Training
• Active Image Synthesis for Efficient Labeling
• Temporal Output Discrepancy for Loss Estimation-based Active Learning
• Stateful Detection of Model Extraction Attacks
• DeepAL: Deep Active Learning in Python
• Diffusion-based Deep Active Learning
• Deep Multi-Fidelity Active Learning of High-dimensional Outputs
• Towards Balanced Active Learning for Multimodal Classification
• Active learning for object detection in high-resolution satellite images
• Multiple-criteria Based Active Learning with Fixed-size Determinantal Point Processes
• High-contrast "gaudy" images improve the training of deep neural network models of visual cortex
• Neural Networks Are More Productive Teachers Than Human Raters: Active Mixup for Data-Efficient Knowledge Distillation from a Blackbox Model
• Effective Evaluation of Deep Active Learning on Image Classification Tasks
• Active Learning Under Malicious Mislabeling and Poisoning Attacks
• Active Deep Learning on Entity Resolution by Risk Sampling
• Sampling Bias in Deep Active Classification: An Empirical Study
• Information Condensing Active Learning
• Selective sampling for accelerating training of deep neural networks
• Ask-n-Learn: Active Learning via Reliable Gradient Representations for Image Classification
• Black-Box Dissector: Towards Erasing-based Hard-Label Model Stealing Attack
• Adaptive Deep Learning for Entity Resolution by Risk Analysis
• Effective Version Space Reduction for Convolutional Neural Networks
• S${}^3$VAADA: Submodular Subset Selection for Virtual Adversarial Active Domain Adaptation
• Active Sentence Learning by Adversarial Uncertainty Sampling in Discrete Space
• VirAAL: Virtual Adversarial Active Learning For NLU
• Exploring Adversarial Examples for Efficient Active Learning in Machine Learning Classifiers
• State-Relabeling Adversarial Active Learning
• Deep Active Learning by Model Interpretability
• Bias-Aware Heapified Policy for Active Learning
• Knowledge-driven Active Learning
• Low-Regret Active learning
• Leveraging Motion Priors in Videos for Improving Human Segmentation
• A Survey on Green Deep Learning
• Mitigating Sampling Bias and Improving Robustness in Active Learning