Research map / Label-efficient and robust learning / Recognition and long-tail
An end-to-end approach for the verification problem: learning the right distance
Joao Monteiro, Isabela Albuquerque, Jahangir Alam, R Devon Hjelm, Tiago Falk
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 the learned pairwise score instead of cosine scoring | Proposed verification model · Cifar-10 | EER: 3.56% -> 3.43% | Improved |
| use the learned pairwise score instead of cosine scoring | Proposed verification model · Mini-ImageNet (Validation) | EER: 30.66% -> 28.64% | Improved |
| use the learned pairwise score instead of cosine scoring | Proposed verification model · Mini-ImageNet (Test) | EER: 32.97% -> 29.26% | Improved |
| add cosine score fusion to E2E score | Proposed verification model · Mini-ImageNet (Test) | EER: 29.26% -> 29.32% | No clear effect |
| add auxiliary cross-entropy loss (ablation) | Proposed speaker verification model · VoxCeleb1 Test set | EER: 3.44% -> 2.51% | Improved |
| add auxiliary cross-entropy loss (ablation) | Proposed speaker verification model · VoxCeleb1-E | EER: 3.70% -> 2.57% | Improved |
| add auxiliary cross-entropy loss (ablation) | Proposed speaker verification model · VoxCeleb1-H | EER: 7.76% -> 4.73% | Improved |
| use learned E2E scoring instead of PLDA | Proposed speaker verification model · VoxCeleb1 Test set | EER: 3.75% -> 2.51% | Improved |
| fine-tune with multilingual data including target-domain speech | Proposed speaker verification model · NIST SRE 2018 CTS | EER: 13.61% -> 8.43% | Improved |
| add auxiliary loss (ablation) | E2E image clustering model · CUB | NMI: 58.5% -> 60.5% | Improved |
| add auxiliary loss (ablation) | E2E image clustering model · CARS | NMI: 60.4% -> 59.9% | Hurt |
| add auxiliary loss (ablation) | E2E image clustering model · SOP | NMI: 88.1% -> 89.2% | Improved |
Closest papers by meaning
- Distance-Based Learning from Errors for Confidence Calibration (ICLR 2020 · similarity 0.86)
- Linear Distance Metric Learning with Noisy Labels (JMLR 2024 · similarity 0.85)
- Asymmetric Metric Learning for Knowledge Transfer (CVPR 2021 · similarity 0.85)
- No Fuss Distance Metric Learning Using Proxies (ICCV 2017 · similarity 0.85)
- Learning Distance for Sequences by Learning a Ground Metric (ICML 2019 · similarity 0.84)
- Stabilizing Sample Similarity in Representation via Mitigating Random Consistency (ICML 2025 · 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.
