Research map / Probabilistic and scientific modeling / Uncertainty and Gaussian processes
Bayesian Estimate of Mean Proper Scores for Diversity-Enhanced Active Learning
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 |
|---|---|---|---|
| use diversity-aware k-means batch selection instead of top-K ranking (ablation) | CoreMSE and CoreLog · PubMed and SST5 text classification | weighted F1 | Improved |
| represent samples by vectors of expected score changes instead of pretrained LM embeddings for clustering (ablation) | CoreMSE and CoreLog · PubMed and SST5 | weighted F1 | Improved |
| use batch acquisition instead of non-batch (B=1) (ablation) | CoreMSE and CoreLog · PubMed and SST5 | weighted F1 | Improved |
| use smaller batch size (5 or 10) instead of 50 or 100 (ablation) | batch CoreMSE · PubMed and SST5 | weighted F1 | Improved |
| use Dynamic validation set for ensemble training (ablation) | batch CoreMSE · PubMed and SST5 | weighted F1 | Improved |
| train 30 epochs without validation set instead of Dynamic VS (ablation) | batch CoreMSE · PubMed and SST5 | weighted F1 | Mixed |
| replace deep ensemble with MC-dropout (ablation) | CoreMSE · SST5 (batch 50, 500 labels) | weighted F1: 0.385 -> 0.400 | No clear effect |
| replace deep ensemble with MC-dropout (ablation) | CoreMSE · SST5 training time | training time (s): 45 -> 9 | Improved |
| increase ensemble size from 5 to 40 (ablation) | CoreMSE · SST5 (batch 50, 500 labels) | weighted F1: 0.385 -> 0.403 | Improved |
| use CoreMSE instead of BADGE for acquisition (ablation) | DistilBERT AL · AG NEWS query time | query time (s): 858 -> 320 | Improved |
| use CoreMSE and CoreLog instead of WMOCU, BADGE and ALPS in batch AL | DistilBERT AL · four text datasets (batch 50) | weighted F1 | Improved |
Closest papers by meaning
- Diversity Enhanced Active Learning with Strictly Proper Scoring Rules (NeurIPS 2021 · similarity 0.92)
- On the Convergence of Loss and Uncertainty-based Active Learning Algorithms (NeurIPS 2024 · similarity 0.86)
- Sharpness-diversity tradeoff: improving flat ensembles with SharpBalance (NeurIPS 2024 · similarity 0.85)
- Uncertainty Herding: One Active Learning Method for All Label Budgets (ICLR 2025 · similarity 0.85)
- Class-Attribute Priors: Adapting Optimization to Heterogeneity and Fairness Objective (AAAI 2024 · similarity 0.85)
- Provably Neural Active Learning Succeeds via Prioritizing Perplexing Samples (ICML 2024 · similarity 0.84)
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.
