Research map / Label-efficient and robust learning / Semi-supervised learning
A Unified Approach to Count-Based Weakly Supervised Learning
Vinay Shukla, Zhe Zeng, Kareem Ahmed, Guy Van den Broeck
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 exact count loss instead of proportion loss PL | LLP · Adult [0,1] bag size 8 | test AUC: 0.8884 -> 0.8985 | Improved |
| use count loss instead of LMMCM | LLP · Magic [0,1] bag size 8 | test AUC: 0.9070 -> 0.9173 | Improved |
| use count loss instead of PL | LLP · Magic [0,1] bag size 512 | test AUC: 0.8807 -> 0.9088 | Improved |
| use count loss at large bag size | LLP · Adult [0,1] bag size 128 | test AUC: 0.8879 -> 0.8871 | No clear effect |
| use count loss instead of attention baselines at small training set | MIL · MNIST N(10,2), 50 bags | test AUC: 0.775 -> 0.818 | Improved |
| use count loss with many training bags | MIL · MNIST N(50,10), 150 bags | test AUC: 0.970 -> 0.964 | Hurt |
| use count loss instead of attention baselines | MIL · Colon Cancer | accuracy: 0.909 -> 0.915 | Improved |
| use count loss instead of Instance-Max | MIL instance-level · MNIST N(10,2), 50 bags | AUC: 0.4904 -> 0.8341 | Improved |
| use CL instead of CVIR | PU learning · Binarized CIFAR | accuracy: 82.3 -> 80.1 | Hurt |
| use CL-expect instead of CVIR | PU learning · CIFAR Cat vs. Dog | accuracy: 73.3 -> 76.5 | Improved |
| use full distribution CL rather than expectation CL-expect (ablation) | PU learning · CIFAR Cat vs. Dog | accuracy: 76.5 -> 74.8 | Hurt |
| use full distribution CL rather than CL-expect (ablation) | PU learning · MNIST17 | accuracy: 98.7 -> 99.0 | Improved |
Closest papers by meaning
- Losses over Labels: Weakly Supervised Learning via Direct Loss Construction (AAAI 2023 · similarity 0.90)
- Loss factorization, weakly supervised learning and label noise robustness (ICML 2016 · similarity 0.88)
- Labels, Information, and Computation: Efficient Learning Using Sufficient Labels (JMLR 2023 · similarity 0.88)
- Lower-Bounded Proper Losses for Weakly Supervised Classification (ICML 2021 · similarity 0.88)
- Learning from weak labelers as constraints (ICLR 2025 · similarity 0.88)
- Forming Auxiliary High-confident Instance-level Loss to Promote Learning from Label Proportions (CVPR 2025 · similarity 0.85)
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.
