Deep Batch Active Learning by Diverse, Uncertain Gradient Lower Bounds

Abstract

We design a new algorithm for batch active learning with deep neural network models. Our algorithm, Batch Active learning by Diverse Gradient Embeddings (BADGE), samples groups of points that are disparate and high-magnitude when represented in a hallucinated gradient space, a strategy designed to incorporate both predictive uncertainty and sample diversity into every selected batch. Crucially, BADGE trades off between diversity and uncertainty without requiring any hand-tuned hyperparameters. We show that while other approaches sometimes succeed for particular batch sizes or architectures, BADGE consistently performs as well or better, making it a versatile option for practical active learning problems.

References in corpus (9)

• Determinantal point processes for machine learning
• Adversarial Active Learning for Deep Networks: a Margin Based Approach
• Agnostic Active Learning Without Constraints
• Deep Active Learning over the Long Tail
• Active Learning for Speech Recognition: the Power of Gradients
• libact: Pool-based Active Learning in Python
• Fast determinantal point processes via distortion-free intermediate sampling
• Uncertainty Sampling is Preconditioned Stochastic Gradient Descent on Zero-One Loss
• Local Uncertainty Sampling for Large-Scale Multi-Class Logistic Regression

Cited by in corpus (33)

• Fine-Tuning Language Models from Human Preferences
• Active learning for data streams: a survey
• A Survey of Active Learning for Text Classification using Deep Neural Networks
• On Warm-Starting Neural Network Training
• DeepGD: A Multi-Objective Black-Box Test Selection Approach for Deep Neural Networks
• Deep Indexed Active Learning for Matching Heterogeneous Entity Representations
• Deep Active Learning for Computer Vision: Past and Future
• MUS-CDB: Mixed Uncertainty Sampling with Class Distribution Balancing for Active Annotation in Aerial Object Detection
• SIMILAR: Submodular Information Measures Based Active Learning In Realistic Scenarios
• Adversarial Representation Active Learning
• Exploring Active Learning for Semiconductor Defect Segmentation
• Deep Active Learning: Unified and Principled Method for Query and Training
• Small-Text: Active Learning for Text Classification in Python
• Active Learning to Classify Macromolecular Structures in situ for Less Supervision in Cryo-Electron Tomography
• Deep Active Learning for Text Classification with Diverse Interpretations
• Placing Human Animations into 3D Scenes by Learning Interaction- and Geometry-Driven Keyframes
• Thoughtful Adoption of NLP for Civic Participation: Understanding Differences Among Policymakers
• GeneDisco: A Benchmark for Experimental Design in Drug Discovery
• Forgetful Active Learning with Switch Events: Efficient Sampling for Out-of-Distribution Data
• Medical symptom recognition from patient text: An active learning approach for long-tailed multilabel distributions
• Towards Balanced Active Learning for Multimodal Classification
• 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
• One-Round Active Learning
• Information Condensing Active Learning
• Ask-n-Learn: Active Learning via Reliable Gradient Representations for Image Classification
• Neural Active Learning with Performance Guarantees
• Zero-Round Active Learning
• An Interactive Visualization Tool for Understanding Active Learning
• Reducing the Long Tail Losses in Scientific Emulations with Active Learning
• Knowledge-driven Active Learning
• Low-Regret Active learning
• Active learning of potential-energy surfaces of weakly-bound complexes with regression-tree ensembles