Research map / Label-efficient and robust learning / Semi-supervised learning
Cost-Sensitive Self-Training for Optimizing Non-Decomposable Metrics
Harsh Rangwani, Shrinivas Ramasubramanian, Sho Takemori, Kato Takashi, Yuhei Umeda, Venkatesh Babu Radhakrishnan
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 CSST's cost-sensitive consistency training to FixMatch | FixMatch · CIFAR10-LT, ρ=100 | Avg. Rec: 0.78 -> 0.80 | Improved |
| Add CSST's cost-sensitive consistency training to FixMatch | FixMatch · CIFAR100-LT, ρ=10 | Min. HT Cov: 0.004 -> 0.010 | Improved |
| Add CSST's cost-sensitive consistency training to FixMatch | FixMatch · ImageNet100-LT, ρ=10 | Min. HT Cov: 0.007 -> 0.010 | Improved |
| Add CSST's cost-sensitive consistency training to UDA | UDA · IMDb, ρ=10 | Min Rec: 0.66 -> 0.88 | Improved |
| Add CSST's cost-sensitive consistency training to UDA | UDA · IMDb, ρ=100 | Min Rec: 0.00 -> 0.75 | Improved |
| Add CSST's cost-sensitive consistency training to UDA | UDA · DBpedia-14, ρ=100 | Min Rec: 0.65 -> 0.97 | Improved |
| Remove KL-based thresholding (ablation) | CSST(FixMatch) · CIFAR10-LT, ρ=100 | Avg. Rec: 0.80 -> 0.67 | Hurt |
| Remove KL-based thresholding (ablation) | CSST(FixMatch) · CIFAR100-LT, ρ=10 | Avg. Rec: 0.63 -> 0.47 | Hurt |
| Remove KL-based thresholding (ablation) | CSST(FixMatch) · ImageNet100-LT, ρ=10 | Avg. Rec: 0.58 -> 0.26 | Hurt |
| Remove KL-based thresholding (ablation) | CSST(FixMatch) · CIFAR10-LT, ρ=100 | Min. Cov: 0.092 -> 0.093 | Mixed |
| Increase the KL-divergence threshold too far (ablation) | CSST(FixMatch) · CIFAR10-LT, ρ=100 | Avg. Rec | Hurt |
| Increase the amount of labeled data while holding unlabeled data constant (ablation) | CSST · Experiments in Figure [ref] | Avg. Rec | Improved |
Closest papers by meaning
- Debiased Self-Training for Semi-Supervised Learning (NeurIPS 2022 · similarity 0.88)
- Theoretical Analysis of Self-Training with Deep Networks on Unlabeled Data (ICLR 2021 · similarity 0.87)
- Analyzing the Sample Complexity of Self-Supervised Image Reconstruction Methods (NeurIPS 2023 · similarity 0.85)
- A Survey on Self-Supervised Learning: Algorithms, Applications, and Future Trends (TPAMI 2024 · similarity 0.85)
- Self-Taught Metric Learning Without Labels (CVPR 2022 · similarity 0.85)
- Doubly-Robust Self-Training (NeurIPS 2023 · 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.
