Hybrid Active Learning with Uncertainty-Weighted Embeddings
Abstract: We introduce a hybrid active learning method that simultaneously considers uncertainty and diversity for sample selection. Our method consists of two key steps: computing a novel uncertainty-weighted embedding, then applying distance-based sampling for sample selection. Our proposed uncertainty-weighted embedding is computed by weighting a sample's feature representation by an uncertainty measure. We show how this embedding generalizes the gradient embedding of BADGE so it can be used with arbitrary loss functions and be computed more efficiently, especially for dense prediction tasks and network architectures with large numbers of parameters in the final layer. We extensively evaluate the proposed hybrid active learning method on image classification, semantic segmentation and object detection tasks, and demonstrate that it achieves state-of-the-art performance.
Submission Length: Regular submission (no more than 12 pages of main content)
Changes Since Last Submission: * Add experiments to validate UWE's capability to approximate real gradient in Section 5.4
* Provide additional results on object detection to reveal BADGE's weaknesses in A.3
* Compare the uncertainty and diversity of samples selected by different active learning methods in A.4
* Provide results on additional dataset mini-Imagenet in A.5
* Provide visualization of detailed detection results in A.6
* Provide raw data for Figures 3, 4, and 5 in A.7
* Fix the citation format problem and modify the first paragraph of Section 4 to provide more computation details of PPM.
Assigned Action Editor: ~Mingsheng_Long2
Submission Number: 2088
Loading