Research map / Learning theory and algorithms / Prediction and decision losses
A Theoretically Sound Upper Bound on the Triplet Loss for Improving the Efficiency of Deep Distance Metric Learning
Check 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 |
|---|---|---|---|
| replace triplet+smart mining with the discriminative loss | GoogLeNet DML · CUB-200-2011 | training time (minutes): 680 -> 54 | Helped |
| replace triplet+smart mining with the discriminative loss | GoogLeNet DML · CAR196 | training time (minutes): 1240 -> 73 | Helped |
| use the discriminative loss instead of triplet+smart mining | GoogLeNet DML · CUB-200-2011 | R@1: 49.78 -> 51.43 | Helped |
| use the discriminative loss instead of triplet+proxy | GoogLeNet DML · CAR196 | NMI: 64.90 -> 59.71 | Hurt |
| use the discriminative loss instead of triplet+proxy | GoogLeNet DML · CAR196 | R@1: 73.22 -> 68.31 | Hurt |
| use the discriminative loss instead of triplet+proxy | GoogLeNet DML · CAR196 | R@8: 88.68 -> 91.18 | Helped |
| use second-to-last rather than last-layer embeddings (ablation) | Discriminative with GoogLeNet · CUB-200-2011 | R@1: 49.49 -> 51.43 | Helped |
| use second-to-last rather than last-layer embeddings (ablation) | Discriminative with GoogLeNet · CAR196 | R@1: 65.32 -> 68.31 | Helped |
| use one-hot rather than K-means centroids (ablation) | Discriminative loss · CUB-200-2011 | R@1: 50.75 -> 51.43 | No clear effect |
| use one-hot rather than K-means centroids (ablation) | Discriminative loss · CAR196 | R@1: 66.93 -> 68.31 | Helped |
| replace GoogLeNet with VGG-16 | Discriminative loss · CUB-200-2011 | R@1: 51.43 -> 57.74 | Helped |
| replace GoogLeNet with VGG-16 | Discriminative loss · CAR196 | R@1: 68.31 -> 78.15 | Helped |
Closest papers by meaning
- Triplet-Center Loss for Multi-View 3D Object Retrieval (CVPR 2018 · similarity 0.87)
- Improved Deep Metric Learning with Multi-class N-pair Loss Objective (NeurIPS 2016 · similarity 0.87)
- Threshold-Consistent Margin Loss for Open-World Deep Metric Learning (ICLR 2024 · similarity 0.86)
- Mathematical Justification of Hard Negative Mining via Isometric Approximation Theorem (ICLR 2024 · similarity 0.85)
- MetricEmbedding: Accelerate Metric Nearness by Tropical Inner Product (ICML 2025 · similarity 0.85)
- Collapse-Aware Triplet Decoupling for Adversarially Robust Image Retrieval (ICML 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.