Research map / Label-efficient and robust learning / Recognition and long-tail
Query-Driven Multi-Instance Learning
Yen-Chi Hsu, Cheng-Yao Hong, Ming-Sui Lee, Tyng-Luh Liu
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 |
|---|---|---|---|
| add attention regularization L2 (1 query) (ablation) | qMIL · CIFAR10, 1 query | accuracy: 82.4 ± 1.7 -> 89.9 ± 1.7 | No clear effect |
| add attention regularization L2 (3 queries) (ablation) | qMIL · CIFAR10, 3 queries | accuracy: 81.22 ± 1.8 -> 81.77 ± 1.4 | No clear effect |
| add attention regularization L2 (5 queries) (ablation) | qMIL · CIFAR10, 5 queries | accuracy: 71.23 ± 3.4 -> 79.45 ± 2.7 | Improved |
| add attention regularization L2 (7 queries) (ablation) | qMIL · CIFAR10, 7 queries | accuracy: 65.66 ± 4.6 -> 82.09 ± 2.1 | Improved |
| add attention regularization L2 (10 queries) (ablation) | qMIL · CIFAR10, 10 queries | accuracy: 78.33 ± 2.3 -> 86.14 ± 1.3 | Improved |
| add attention regularization L2 (10 queries) (ablation) | qMIL · CIFAR10, 10 queries | attention accuracy: 70.64 ± 5.4 -> 89.18 ± 1.0 | Improved |
| add attention regularization L2 (7 queries) (ablation) | qMIL · CIFAR10, 7 queries | attention accuracy: 45.66 ± 20.3 -> 86.01 ± 1.2 | Improved |
| add attention regularization L2 (3 queries) (ablation) | qMIL · CIFAR10, 3 queries | attention accuracy: 65.52 ± 9.9 -> 87.22 ± 1.1 | Improved |
| replace GatedAttnDMIL with qMIL (5000 training bags) | qMIL · CIFAR10 single query | accuracy: 82.4 ± 1.7 -> 89.9 ± 1.7 | Improved |
| replace GatedAttnDMIL with qMIL (100 training bags) | qMIL · CIFAR10 single query | accuracy: 55.1 ± 8.6 -> 56.3 ± 4.5 | No clear effect |
| increase training bags from 100 to 5000 | qMIL · CIFAR10 single query | accuracy: 56.3 ± 4.5 -> 89.9 ± 1.7 | Improved |
| replace deep MIML with qMIL | qMIL · MIML Scene | accuracy: 89.45 ± 1.22 -> 90.20 ± 0.96 | Improved |
Closest papers by meaning
- Deep MIML Network (AAAI 2017 · similarity 0.86)
- Discover Multiple Novel Labels in Multi-Instance Multi-Label Learning (AAAI 2017 · similarity 0.86)
- Fast Multi-Instance Multi-Label Learning (TPAMI 2018 · similarity 0.85)
- MIML-FCN+: Multi-Instance Multi-Label Learning via Fully Convolutional Networks With Privileged Information (CVPR 2017 · similarity 0.85)
- Disambiguated Attention Embedding for Multi-Instance Partial-Label Learning (NeurIPS 2023 · similarity 0.84)
- Self-Paced Unified Representation Learning for Hierarchical Multi-Label Classification (AAAI 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.
