Research map / Label-efficient and robust learning / Recognition and long-tail
SoDeep: A Sorting Deep Net to Learn Ranking Loss Surrogates
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 MSE loss with SoDeep loss | Sem-Emb memorability model · Media Memorability test | Spearman correlation: 48.6 -> 49.4 | Helped |
| replace MSE loss with SoDeep loss | R34 memorability model · Media Memorability test | Spearman correlation: 44.2 -> 46.6 | Helped |
| add SoDeep mAP loss to the soft margin loss | WILDCAT* · VOC 2007 | mAP: 93.2 -> 94.0 | Helped |
| replace triplet margin loss with the SoDeep rank-based loss | DSVE-Loc · MS-COCO rVal caption retrieval | R@1: 69.8 -> 71.5 | Helped |
| replace triplet margin loss with the SoDeep rank-based loss | DSVE-Loc · MS-COCO rVal caption retrieval | R@5: 91.9 -> 92.8 | Helped |
| replace triplet margin loss with the SoDeep rank-based loss | DSVE-Loc · MS-COCO rVal image retrieval | R@1: 55.9 -> 56.2 | Helped |
| replace the LSTM sorter with a CNN sorter (ablation) | SoDeep sorter · Synthetic sorting test set | L1 loss: 0.0033 -> 0.0120 | Hurt |
| replace the LSTM sorter with the handcrafted pairwise-comparison sorter | SoDeep sorter · Synthetic sorting test set | L1 loss: 0.0033 -> 0.0350 | Hurt |
| use 8 or more convolutional layers (ablation) | CNN sorter · Synthetic sorting | cost function | No clear effect |
| use the direct differentiable Recall@K loss (ablation) | SoDeep retrieval training · Cross-modal retrieval | recall | No clear effect |
Closest papers by meaning
- PiRank: Scalable Learning To Rank via Differentiable Sorting (NeurIPS 2021 · similarity 0.87)
- RankMI: A Mutual Information Maximizing Ranking Loss (CVPR 2020 · similarity 0.86)
- RankDNN: Learning to Rank for Few-Shot Learning (AAAI 2023 · similarity 0.86)
- Differentiable Sorting Networks for Scalable Sorting and Ranking Supervision (ICML 2021 · similarity 0.85)
- Deep Metric Learning to Rank (CVPR 2019 · similarity 0.85)
- ST$_k$: A Scalable Module for Solving Top-k Problems (NeurIPS 2024 · similarity 0.83)
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.