Research map / Label-efficient and robust learning / Recognition and long-tail
Beam Search Optimized Batch Bayesian Active Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 diverse beam search with a self-adapted constraint instead of greedy BatchBALD | batch active learning · MNIST, training set size 200 | accuracy: 0.941±0.008 -> 0.948±0.006 | Helped |
| use diverse beam search with a self-adapted constraint instead of greedy BatchBALD | batch active learning · ReMNIST, training set size 200 | accuracy: 0.924±0.007 -> 0.939±0.005 | Helped |
| use diverse beam search with a self-adapted constraint instead of greedy BatchBALD | batch active learning · CIFAR-10, training set size 200 | accuracy: 0.301±0.011 -> 0.309±0.012 | Helped |
| add the self-adapted constraint to beam search (ablation) | beam-search active learning · ReMNIST, training set size 150 | accuracy: 0.918 ± 0.006 -> 0.927 ± 0.005 | Helped |
| add the self-adapted constraint to beam search (ablation) | beam-search active learning · MNIST, training set size 150 | accuracy: 0.928 ± 0.004 -> 0.931 ± 0.004 | Helped |
| add the self-adapted constraint to beam search (ablation) | beam-search active learning · CIFAR-10, training set size 150 | accuracy: 0.289 ± 0.006 -> 0.306 ± 0.005 | Helped |
| increase beam width (ablation) | beam-search active learning · ReMNIST | accuracy | Helped |
| increase beam width beyond 32 (ablation) | beam-search active learning · ReMNIST | accuracy | No clear effect |
| set beam-diversity parameter γ to values in (0.5, 1.2) (ablation) | beam-search active learning · ReMNIST, training set size 100 | accuracy: 0.897 ± 0.007; 0.897 ± 0.006; 0.897 ± 0.008; 0.893 ± 0.006; 0.894 ± 0.005; 0.893 ± 0.006 | No clear effect |
| set beam-diversity parameter γ to 100 (ablation) | beam-search active learning · ReMNIST, training set size 100 | accuracy: 0.897 ± 0.007 -> 0.601 ± 0.006; 0.893 ± 0.006 -> 0.506 ± 0.005 | Hurt |
Closest papers by meaning
- Scalable Batch-Mode Deep Bayesian Active Learning via Equivalence Class Annealing (ICLR 2023 · similarity 0.89)
- Batch Active Learning at Scale (NeurIPS 2021 · similarity 0.88)
- Bayesian Batch Active Learning as Sparse Subset Approximation (NeurIPS 2019 · similarity 0.88)
- BatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active Learning (NeurIPS 2019 · similarity 0.88)
- Querying Easily Flip-flopped Samples for Deep Active Learning (ICLR 2024 · similarity 0.84)
- Training Greedy Policy for Proposal Batch Selection in Expensive Multi-Objective Combinatorial Optimization (ICML 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.