Research map / Learning theory and algorithms / Prediction and decision losses
Rethinking the Optimization of Average Precision: Only Penalizing Negative Instances before Positive Ones Is Enough
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 PNP-Dq instead of PNP-O's constant gradient assignment (ablation) | PNP loss · SOP | R@1: 77.9 -> 80.1 | Helped |
| Use an increasing unbounded derivative rather than PNP-O (ablation) | PNP loss · SOP | R@1: 77.9 -> 76.3 | Hurt |
| Use an increasing bounded derivative rather than PNP-O (ablation) | PNP loss · SOP | R@1: 77.9 -> 77.9 | No clear effect |
| Use a decreasing slow derivative rather than PNP-O (ablation) | PNP loss · SOP | R@1: 77.9 -> 79.6 | Helped |
| Use PNP-Dq instead of PNP-O's constant gradient assignment (ablation) | PNP loss · VehicleID small | R@1: 93.6 -> 94.6 | Helped |
| Use an increasing unbounded derivative rather than PNP-O (ablation) | PNP loss · VehicleID small | R@1: 93.6 -> 92.4 | Hurt |
| Use an increasing bounded derivative rather than PNP-O (ablation) | PNP loss · VehicleID small | R@1: 93.6 -> 94.0 | Helped |
| Use a decreasing slow derivative rather than PNP-O (ablation) | PNP loss · VehicleID small | R@1: 93.6 -> 94.3 | Helped |
| Increase batch size from 112 to 384 | PNP-Dq · SOP | R@1: 80.1 -> 81.1 | Helped |
| Increase batch size from 112 to 384 | PNP-Dq · VehicleID small | R@1: 94.6 -> 95.5 | Helped |
| Use PNP-Dq rather than Smooth-AP under merged-class, multi-center training | Retrieval training · VehicleID small, robustness evaluation | R@1: +2 - 3% | Helped |
| Increase the PNP-Dq steepness parameter alpha (ablation) | PNP-Dq · SOP and VehicleID | R@1 | Helped |
Closest papers by meaning
- QuRe: Query-Relevant Retrieval through Hard Negative Sampling in Composed Image Retrieval (ICML 2025 · similarity 0.84)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 2024 · similarity 0.83)
- Composed Image Retrieval with Text Feedback via Multi-grained Uncertainty Regularization (ICLR 2024 · similarity 0.83)
- Generalized Gradient Norm Clipping & Non-Euclidean $(L_0,L_1)$-Smoothness (NeurIPS 2025 · similarity 0.82)
- Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval (ICLR 2025 · similarity 0.82)
- A New Robust Partial p-Wasserstein-Based Metric for Comparing Distributions (ICML 2024 · similarity 0.82)
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.