Research map / Learning theory and algorithms / Prediction and decision losses
Diversity Enhanced Active Learning with Strictly Proper Scoring Rules
Wei Tan, Lan Du, Wray Buntine
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| batch acquisition with k-Means diversity instead of top-B scores (CoreMSE_top) (ablation) | batch CoreMSE · PUBMED/SST-5/AG NEWS/IMDB | F1 learning curve | Improved |
| batch acquisition with k-Means diversity instead of top-B scores (CoreLog_top) (ablation) | batch CoreLog · PUBMED/SST-5/AG NEWS/IMDB | F1 learning curve | Improved |
| cluster expected-score-change vectors instead of DistilBERT last-layer embeddings (Pretrained_LM) (ablation) | batch CoreMSE/CoreLog · text classification datasets | learning curve | Improved |
| use batch of size 5 or 10 instead of 50 or 100 | batch CoreMSE · PUBMED, SST-5 | learning curve | Improved |
| use batch of size 5 or 10 instead of no-batch (B=1) | CoreMSE · PUBMED, SST-5 | learning curve | Improved |
| dynamic validation set versus 30 epochs without validation set (ablation) | batch CoreMSE · PUBMED, SST-5, AG NEWS | learning curve | Mixed |
| dynamic validation set versus fixed-length (1000) validation set (ablation) | batch CoreMSE · PUBMED, SST-5, AG NEWS | learning curve | Improved |
| dynamic validation set versus constant validation set shared across ensemble (ablation) | batch CoreMSE · PUBMED, SST-5, AG NEWS | learning curve | Improved |
| use BEMPS CoreMSE instead of WMOCU/Max-Entropy/BALD (batch size one) | active learning · SST-5, PUBMED | weighted F1 | Improved |
| use batch WMOCU with our batch algorithm versus BADGE and ALPS | batch WMOCU · four text datasets | learning curve | Improved |
| increase unlabelled pool from 10k to 100k | BADGE · AG NEWS | running time (s): 76 -> 2904 | Hurt |
| increase unlabelled pool from 10k to 100k | CoreMSE · AG NEWS | running time (s): 65 -> 654 | No clear effect |
Closest papers by meaning
- Bayesian Estimate of Mean Proper Scores for Diversity-Enhanced Active Learning (TPAMI 2023 · similarity 0.92)
- Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples (ICML 2024 · similarity 0.84)
- Universal Rates for Active Learning (NeurIPS 2024 · similarity 0.84)
- Scaling Up Active Testing to Large Language Models (NeurIPS 2025 · similarity 0.84)
- Uncertainty Herding: One Active Learning Method for All Label Budgets (ICLR 2025 · similarity 0.83)
- On the Convergence of Loss and Uncertainty-based Active Learning Algorithms (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
