Research map / Label-efficient and robust learning / Recognition and long-tail
Mitigating Negative Flips via Margin Preserving Training
Simone Ricci, Niccolò Biondi, Federico Pernici, Alberto Del Bimbo
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 margin-calibrating logit bias (with KL distillation) (ablation) | MPT-KL · ImageNet-1k (ResNet-18 to ResNet-18) | NFR (%): 8.09 -> 6.09 | Improved |
| add double-source focal distillation (with bias) (ablation) | MPT · ImageNet-1k (ResNet-18 to ResNet-18) | NFR (%): 7.71 -> 6.09 | Improved |
| add logit bias alone (no distillation) (ablation) | cross-entropy training · ImageNet-1k (ResNet-18 to ResNet-18) | ER on Yall (%): 32.88 -> 34.13 | Hurt |
| add logit bias alone (no distillation) (ablation) | cross-entropy training · ImageNet-1k (ResNet-18 to ResNet-18) | ER on Yold (%): 28.40 -> 26.23 | Improved |
| increase margin k on new classes (ablation) | MPT · CIFAR100 | NFR | Improved |
| use MPT-KL instead of ELODI ensemble of eight new models | model update · ImageNet-1k (ResNet-18 to ResNet-18) | NFR (%): 7.61 -> 6.09 | Improved |
| use MPT-LM instead of ELODI | model update · CIFAR100 (ResNet-18 to ResNet-50) | NFR (%): 9.10 -> 9.42 | Hurt |
| apply MPT to pretrained ViT-B/32 classifier fine-tuning | ViT-B/32 · CIFAR100 (pretrained ViT-B/32, classifier fine-tuned) | NFR (%): 7.08 -> 3.26 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ELODI — Elodi: Ensemble Logit Difference Inhibition for Positive-Congruent Training (TPAMI 2024) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Positive-Congruent Training: Towards Regression-Free Model Updates (CVPR 2021 · similarity 0.88)
- Elodi: Ensemble Logit Difference Inhibition for Positive-Congruent Training (TPAMI 2024 · similarity 0.86)
- Stable Coresets via Posterior Sampling: Aligning Induced and Full Loss Landscapes (NeurIPS 2025 · similarity 0.84)
- Decoupled Kullback-Leibler Divergence Loss (NeurIPS 2024 · similarity 0.83)
- MaxSup: Overcoming Representation Collapse in Label Smoothing (NeurIPS 2025 · similarity 0.83)
- Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers (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.
